The Short-Term Impacts of the 2017 Portuguese Wildfires on Human Health and Visibility: A Case Study

Abstract

The frequency of extreme wildfire events (EWEs) is expected to increase due to climate change, leading to higher levels of atmospheric pollutants being released into the air, which could cause significant short-term impacts on human health (both for the population and firefighters) and on visibility. This study aims to gain a better understanding of the effects of EWEs’ smoke on air quality, its short-term impacts on human health, and how it reduces visibility by applying a modelling system to the Portuguese EWEs of October 2017. The Weather Research and Forecasting Model was combined with a semi-empirical fire spread algorithm (WRF-SFIRE) to simulate particulate matter smoke dispersion and assess its impacts based on up-to-date numerical approaches. Hourly simulated particulate matter values were compared to hourly monitored values, and the WRF-SFIRE system demonstrated accuracy consistent with previous studies, with a correlation coefficient ranging from 0.30 to 0.76 and an RMSE varying between 215 µg/m³ and 418 µg/m³. The estimated daily particle concentration levels exceeded the European air quality limit value, indicating a potential strong impact on human health. Health indicators related to exposure to particles were estimated, and their spatial distribution showed that the highest number of hospital admissions (>300) during the EWE, which occurred downwind of the fire perimeters, were due to the combined effect of high smoke pollution levels and population density. Visibility reached its worst level at night, when dispersion conditions were poorest, with the entire central and northern regions registering poor visibility levels (with a visual range of less than 2 km). This study emphasises the use of numerical models to predict, with high spatial and temporal resolutions, the population that may be exposed to dangerous levels of air pollution caused by ongoing wildfires. It offers valuable information to the public, civil protection agencies, and health organisations to assist in lessening the impact of wildfires on society.

1. Introduction

The frequency of extreme wildfire events (EWEs) is expected to increase due to climate change, which, although a long-term phenomenon, has direct implications for the increase in short-term events such as heat waves and dry spells [1]. Smoke is one of the most disturbing consequences of wildfires, releasing large amounts of pollutants into the atmosphere (e.g., [2]), which strongly impact human health [3,4,5,6] and impair visibility [7,8]. It contains coarse (PM₁₀) and fine (PM_2.5) particulate matter, carbon monoxide (CO), volatile organic compounds (VOCs), and other harmful substances [6,9,10]. The World Health Organization (WHO), aware of the health effects of smoke from wildfires, provided air quality guidelines aimed at protecting populations from smoke [11], particularly in the wildland–urban interface where the risk of human exposure can be higher [12,13].

Exposure to high levels of air pollution during wildfire events can result in a wide range of adverse health consequences, including an increase in respiratory morbidity [14,15,16,17,18]. Moreover, emerging evidence suggests adverse cardiovascular outcomes [19,20,21] due to pollutants’ ability to penetrate deep into the lungs and induce systemic inflammation and oxidate stress [22,23,24,25,26]. This effect could potentially trigger a cascade of pathophysiological events in the human body, leading to various manifestations of coronary heart disease [22,23,27,28,29,30]. The most vulnerable populations to smoke exposure are the common at-risk groups, such as children and older adults [31], as well as personnel involved in firefighting operations [5,6].

In addition to its impact on human health, wildfire smoke can drastically reduce visibility, which poses significant challenges for firefighting efforts and emergency evacuations, consequently endangering suppression activities (e.g., aerial firefighting and wildland firefighter operations) and the safe evacuation of the affected population [7,32].

The 2017 EWEs in Portugal highlighted the critical importance of understanding the impacts of smoke on human health and visibility [33,34,35]. During these EWEs, several air quality monitoring stations registered high concentrations of atmospheric pollutants caused by wildfire emissions and air pollution transport from burning areas [36], and many people needed medical assistance because of smoke intoxication [37,38]. Due to the severity of these EWEs, several studies have been performed by the scientific community regarding smoke emissions [33], climate drivers [1], and air pollution impacts on human health and their associated costs [34,35]. Barbosa et al. [34] assessed the long-term health effects of the 2017 EWEs in Portugal on children and the associated economic burden. A Eulerian chemical transport model was applied over mainland Portugal, with a coarse spatial resolution (0.1° × 0.1°), to obtain the annual averages of PM₁₀ and nitrogen dioxide (NO₂). Moreover, the WHO’s Health Risk of Air Pollution in Europe (HRAPIE) approach, based on health economic assessment tools, was used to calculate the health burden. Augusto et al. [35] analysed the associations between air pollution and mortality during the Portuguese EWEs in October 2017 by applying Poisson regression models. Ground-level PM₁₀ observations with a coarse spatial representation and daily mortality datasets by district were used.

In this context, one question still needs to be explored: what are the short-term and high spatial resolution impacts of extreme wildfire events on human health and visibility? Therefore, this paper’s main goal is to quantitatively assess, with high spatial and temporal resolutions, the particulate pollution caused by an EWE and its short-term effects on human health (population and firefighters) and on visibility. The October 2017 EWEs in Portugal were chosen as the case study, and the dispersion of emitted PM₁₀ and PM_2.5 was computed using the Weather Research and Forecasting Model in conjunction with a semi-empirical fire spread algorithm (WRF-SFIRE). PM is especially associated with forest fires [39] due to the high level of emissions released into the atmosphere that can be transported over long distances by wind and its impact on human health and visibility [33]. Health indicators, namely hospital admissions, related to exposure to PM_2.5 were estimated, and visibility impairment was calculated based on simulated PM levels.

2. A Description of the Case Study

The year 2017 will go down in history as one of the deadliest and most severe wildfire seasons in Portugal [1,37,40]. On 15 and 16 October, seven EWEs were recorded in the central region of Portugal, resulting in a significant number of human casualties, including 48 fatalities and approximately 67 injured individuals. These events also led to the destruction of 1712 homes and 768 company facilities, burning more than 241,000 hectares in less than 24 h, which is equivalent to about 2.4% of the national territory [37,40]. Additionally, they caused various environmental damages [37,40], such as the transport of fire-generated atmospheric pollutants to Northwest Europe and their impact on populations far from the EWEs [35]. The period of abnormal droughts and heat waves [41,42], an unusual meteorological phenomenon in the Portuguese territory (i.e., Hurricane Ophelia) [1,35], along with partially demobilised firefighter teams (during a period outside the official fire season window established by the Portuguese authorities), were identified as the factors behind hundreds of fire ignitions during these EWEs [37,40].

These seven EWEs were initiated by one or more ignitions in the central region of mainland Portugal, and they spread continuously from 6:00 on 15 October to 8:00 on 16 October, 2017. Figure 1 shows the locations of the areas affected by the seven EWEs as well as the Nomenclature of Territorial Units for Statistics (NUTS) II borders.

The first forest fire alert was recorded in Seia at 6:00, leading to the burning of about 17,000 hectares of forest and shrubland. Three hours later (at 9:00), the second wildfire outbreak occurred in Lousã, resulting in the burning of 54,400 hectares. The EWE in Oliveira do Hospital began at 10:26 and was one of the most devastating, resulting in 25 human fatalities, the extensive destruction of homes and industrial facilities, and a vast area of burned forest (51,430 ha). The EWE in Sertã started around noon on 15 October and was brought under control by protection authorities thirteen hours later following some rainfall (30,977 ha burned). Along the coastal area of mainland Portugal, the first wildfire warning was issued around 14:00 in Leiria and Quiaios, leading to burnt areas of 20,014 and 23,844 ha, respectively. The last fire alert was reported at 17:00 in Vouzela, resulting in the smallest burned area of 15,959 ha [37]. Figure A1 provides the timeline of these EWEs.

3. Material and Methods

This section outlines the approach for assessing the effects of these EWEs, namely to estimate air pollution levels with high spatial and temporal resolutions (Section 3.1) as well as their impacts on human health (Section 3.2) and visibility (Section 3.3). The method used to assess the accuracy of the numerical approach is also presented in Section 3.4. A flowchart summarising the research methodology is provided in Figure A2, which illustrates the sequential steps and processes involved in this study.

Data analysis was conducted using Python, a powerful and versatile programming language widely recognised in the scientific community for its robustness in handling large datasets and complex analyses. Key libraries such as NumPy, Pandas, Matplotlib, and netCDF4 were leveraged to perform comprehensive data processing, visualization, and statistical analysis. This approach enabled the efficient management and analysis of the extensive data generated in this study.

3.1. Air Pollution Modelling Setup

The Weather Research and Forecasting Model (WRF) [43], in conjunction with a semi-empirical fire spread algorithm (SFIRE), was used to compute the dispersion of emitted particles to the atmosphere, particularly those with an equivalent diameter smaller than 10 µm (PM₁₀) and 2.5 µm (PM_2.5), during the EWEs. WRF-SFIRE can resolve meteorology, fuel moisture, fire spread, fire emissions, and smoke dispersion of air pollutants emitted by forest fires over short periods in a fully coupled way [44]. It includes a fuel moisture model that estimates moisture based on local fuel load and meteorological conditions (i.e., temperature, relative humidity, and precipitation). The fire spread model simulates landscape-scale physics through a two-way integration with dynamical thermodynamic meteorology in the atmospheric boundary layer (e.g., fire affected by wind, fire heat, and moisture), utilising data such as ignition points and topography. Using results from the fire spread model and fuel-specific emission factors, smoke emissions were quantified. Subsequently, the dispersion of smoke was assessed based on the outcomes from both the WRF (e.g., wind fields) and fire emissions models.

The WRF-SFIRE system was set up using a coarse domain (D1) with a horizontal resolution of 20 km × 20 km, covering the Iberian Peninsula. Additionally, a nested domain (D2) with a resolution of 4 km × 4 km was used to cover mainland Portugal. To address the seven EWEs, two nested domains per EWE were employed, resulting in a total of 14 nested domains. These domains had spatial resolutions of 1 km × 1 km and 200 m × 200 m, providing greater spatial and temporal detail. Moreover, the innermost domain (200 m × 200 m) of each EWE also included a nested fire grid to simulate fire progression and its emissions into the atmosphere, featuring a refined mesh with a horizontal resolution of 20 m. Given the complexity and fine spatial resolution of the WRF-SFIRE model setup, which already involves multiple nested domains and a high level of detail, the decision was made to prioritise the computational feasibility of the study. Therefore, to balance accuracy and resource constraints, the simulations were initiated without a spin-up period, focusing on the specific events of interest.

The model’s physical configuration was determined based on a sensitivity analysis from previous studies conducted in Portugal [45,46]. The parameterizations included the following: the Morrison double-moment scheme [47]; the Rapid Radiative Transfer Model or the short- and long-wave radiation model (RRTMG) [48]; the MM5 similarity surface layer scheme [49]; the Noah Land Surface Model [50] with soil temperature and moisture in four layers, fractional snow cover, and frozen soil physics; the Mellor–Yamada–Janjic (MYJ) Planetary Boundary Layer scheme [51]; and the Grell–Freitas Ensemble Scheme for cumulus parameterization [52].

To initialise the WRF-SFIRE system, four types of data are necessary: (i) global meteorological fields; (ii) fuel load; (iii) ignition points; and (iv) emission factors. The global meteorological data were sourced from the European Centre for Medium-Range Weather Forecasts (ECMWF) Re-Analysis Interim (ERA-Interim) model, featuring a spatial resolution of 0.75° and a temporal resolution of 6 h for surface and pressure levels [53]. The fuel load data were obtained from the 6th Portuguese Forest Inventory, which were compiled through the collection of data from aerial images and field measurements of vegetation at approximately 12,000 sites across mainland Portugal in 2015. This inventory provides detailed information on the surface fuel load categorised by land use type (such as trees, under vegetable cover, standing trees, fallen trees, stump, and foliage) and by forest species (including acacia, carob tree, chestnut, cork oak, eucalyptus, holm oak, oaks, other hardwoods, other resinous and pinus pinaster, and stone pine) [54]. The ignition sites and times for each EWE were determined by Viegas et al. [37] through a comprehensive analysis that involved field data collection, including interviews with firefighting teams and affected individuals. Emission factors suitable for Southern European conditions were chosen based on a literature review that considered land use types in the central region of Portugal [37,55,56,57,58,59].

3.2. Human Health Impacts

PM_2.5 consists of tiny particles that can easily enter the thoracic region of the respiratory system and reach the bloodstream, similar to oxygen molecules [60]. The negative health effects of PM_2.5 exposure start when these particles are inhaled and penetrate the lungs and bloodstream, leading to respiratory and cardiovascular diseases (CVDs) [61]. Therefore, respiratory diseases and CVDs were selected as key indicators for assessing the short-term health impacts of PM2.5 exposure. Health impacts are often expressed through morbidity and mortality indicators derived from epidemiological studies, being mostly related to respiratory and cardiovascular diseases [60,62,63,64,65]. Epidemiological studies combine meta-analyses conducted during air pollution episodes to provide statistical associations by relating ambient concentration changes and different types of health outcomes [60,66]. The resulting concentration–response functions could then be included in linear or non-linear relative risk models, which may or may not contain threshold exposure values, and they are used to translate unit concentration changes in health impacts.

Short-term (24 h) health effects (Equation (1)) of PM_2.5 exposure were estimated by considering non-linear (Equation (2)) concentration–response functions (CRFs) and relative risks (RRs) provided by the Health Risks of Air Pollution in Europe (HRAPIE) project. This methodology is specifically designed to assess acute health impacts within a 24 h period and not account for lag times beyond this timeframe. This approach was selected to align with the immediate objective of capturing the acute impact of air pollution [67].

$${\mathrm{H}\mathrm{e}\mathrm{a}\mathrm{l}\mathrm{t}\mathrm{h}\mathrm{e}\mathrm{f}\mathrm{f}\mathrm{e}\mathrm{c}\mathrm{t}}_{\mathrm{p}}={\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\mathrm{I}\mathrm{n}\mathrm{c}}_{\mathrm{i},\mathrm{j}}\times {\mathrm{P}\mathrm{o}\mathrm{p}}_{\mathrm{j}}\times {\mathrm{R}\mathrm{R}}_{\mathrm{i},\mathrm{p}}$$

(1)

$$\mathrm{Non}\text{-}\mathrm{linear}:{RR}_{i,p}=exp\left[{\beta }_i\times \ln \left(C_P+1\right)-\ln (C_0+1)\right]$$

(2)

where Inc is the incidence rate of the health endpoint i for a population group, j; Pop is the spatial and temporal distributions of a population group, j; RR correlates a pollutant p concentration variation (C − C₀) with the probability of experiencing or avoiding a specific health endpoint, i; the β coefficient denotes the change in the RR for a unit change in concentration, C (expressed as the natural logarithm of RR); C is the concentration of the pollutant, p (µg/m³); and C₀ indicates the pollutant p cut-off or counterfactual concentration value (µg/m³) above from which health impacts are calculated.

The incidence rate for the analysed health endpoints (i.e., hospital admissions for respiratory diseases and CVDs) by population group in 2017 was acquired from the National Health Service [68]. These data were derived from the quarterly trends in admission rates, length of stay, and mortality rates, categorised by institution and based on chapters 9 (cardiovascular diseases) and 10 (respiratory diseases) of the principal diagnosis according to the International Classification of Diseases (ICD), 9th and 10th Revisions, Clinical Modification (CM), and Procedure Coding System (PCS). The spatial distribution of the population (at the neighbourhood level) was obtained from the Portuguese Census dataset [69] (Figure A3), while the hourly geographical locations of the fire occurrences were used as proxies for the temporal and spatial distributions of the firefighter teams (Figure A3) [37]. The default values of C₀ and β (95% confidence interval—CI) by health endpoints (

Table 1. The input datasets considered to assess the short-term health impacts.

Pollutant	Health Endpoint	C0	β (95% CI) per 10 µg/m3
PM2.5	Hospital admissions, respiratory disease, 0+ years	>25 µg/m3	2.91 × 10−2 (95% CI: 0–6.10 × 10−2)
	Hospital admissions: CVD (including stroke), 0+ years		1.40 × 10−2 (95% CI: 2.63 × 10−3–2.55 × 10−2)

) were provided by the AirQ+ software (tool for health risk assessment of air pollution, v.2.2) [70]. The spatial (4 km × 4 km) and temporal (hourly) resolutions of the PM_2.5 levels were delivered by the WRF-SFIRE modelling system (Section 3.1).

3.3. Visibility Impairment

The visual range or visibility is inversely proportional to the light extinction coefficient [71]. Light extinction mainly occurs due to the absorption and scattering effects of particles (PM₁₀ and PM_2.5) in the air [72], and it can be expressed as the sum of air pollution levels multiplied by its absorption efficiency (Equation (3)) [7].

$$b_{ext}\left(m^{-1}\right)=\sum C_p\times {\alpha }_p$$

(3)

where b_ext is the light extinction coefficient, α_p is the specific scattering or absorption efficiency of the pollutant, p (PM₁₀ = 1 m²/µg and PM_2.5 = 1.2 m²/µg) [73], and C_p represents the spatial and temporal concentrations (by grid cell) of the pollutant, p (i.e., PM₁₀ and PM_2.5) obtained from the WRF-SFIRE results (Section 3.1). This approach incorporates air pollution levels from the first model layer, which extends up to approximately 2000 m above the ground. This height was chosen to reflect the visibility conditions most relevant to both ground-level observers, such as drivers and pedestrians, as well as aerial firefighting operations, where visibility is crucial for safe and effective operations.

The output results are presented in visibility classes (

Table 2. Deciview (dV) values and corresponding visibility classes and visual ranges (km) [77].

Deciviews (dV)	Visibility Classes	Visual Range (km)
≤25 dV	Clear	≥30 km
25 < dV ≤ 36	Moderate	10 ≤ km < 30
36 < dV ≤ 52	Low	2 ≤ km < 10
>52 dV	Poor	<2 km

) based on the estimated visibility index (deciview − dV) from Equation (4) [74]. The visual range presented in Table 2 was quantified using Koschmieder’s formula (Equation (5)), where the constant, K, is equal to 3.912 [72,75,76].

$$dV=10\times \ln \left({\displaystyle \frac{b_{ext}}{0.01}}\right)$$

(4)

$$Visualrange(\mathrm{k}\mathrm{m})={\displaystyle \frac{ConstantK}{b_{ext}}}$$

(5)

3.4. Modelling Evaluation

In Portugal, the Portuguese Environment Agency (APA) manages the air quality monitoring network, which provides hourly data on atmospheric concentrations for critical pollutants [78]. The hourly levels of PM₁₀ and PM_2.5 calculated using the WRF-SFIRE system were compared with the hourly measurements obtained from the Portuguese air quality monitoring network. The methods for measuring and processing PM₁₀ and PM_2.5 data follow the guidelines of the European Ambient Air Quality Directive (2008/50/EC). Accordingly, the reference methods for the sampling and measurement of the PM₁₀ and PM_2.5 values are described in EN 12341:1999, “Air Quality—Determination of the PM₁₀ fraction of suspended particulate matter—Reference method and field test procedure to demonstrate reference equivalence of measurement methods”, and EN 14907:2005, “Standard gravimetric measurement method for the determination of the PM_2.5 mass fraction of suspended particulate matter”, respectively [78].

The system’s performance was evaluated following the methodology outlined by Bossioli et al. [79], and the analysis integrated the following statistical parameters: correlation coefficient (r), mean bias (MB), root mean square error (RMSE), WRF-SFIRE average (WRF-SFIRE), and measured average (Obs). The spatial distribution of air quality monitoring stations across the country is uneven, with a higher spatial density in the most populated areas [39], and therefore, the assessment only included four air quality monitoring stations which were impacted by smoke emissions from the seven EWEs and satisfied the minimum data collection efficiency requirement of 75% (according to the code of practise). From these four monitoring stations, two are suburban background stations (i.e., EST and ILH), one is an urban traffic station (i.e., AVE), and another one is a rural background station (i.e., ERV). The altitudes of their locations range from 14 m to 68 m. Detailed information about each station is presented in

Table A1. Characteristics of air quality monitoring stations, including their ambient environments, influences, altitudes, and geographic coordinates (longitude and latitude).

Station	Ambient Environment	Influence	Altitude (m)	Longitude (°)	Latitude (°)
EST	Suburban	Background	14	−8.56716	40.7586
AVE	Urban	Traffic	16	−8.64801	40.6372
ILH	Suburban	Background	14	−8.67203	40.5909
ERV	Rural	Background	68	−8.89294	39.9246

. Due to technical limitations, the reference equipment can measure a maximum concentration of 1000 µg/m³ for particulate matter.

Moreover, the estimated short-term (24 h) human health impacts were compared with the monthly number of hospital admissions registered by the Portuguese health service. This service provides monthly hospital morbidity and mortality data by institution [80]. Despite the different time periods of the simulated and reported data, the comparison allows us to verify whether the values obtained are consistent with those reported.

4. Results

The results and discussion are based on the air quality simulation and its evaluation (Section 4.1) as well as on the short-term human health impacts (Section 4.2) and visibility impairment (Section 4.3) caused by the PM concentration levels during the Portuguese EWEs in October 2017.

4.1. Air Quality

Figure 2 illustrates the spatial distribution of the daily average PM₁₀ and PM_2.5 concentrations, simulated using the WRF-SFIRE system at a 4 km × 4 km resolution (D2), between 15 and 16 October 2017, over 24 h. Additionally, the daily average measurements from the air quality monitoring stations are presented as small coloured circles, using the same concentration scale for both the modelling and measurement results.

The daily averaged PM₁₀ levels measured by the air quality monitoring stations exceeded the daily limit value (50 µg/m⁻³) set by the European Air Quality Directive (2008/50/EC) at several locations: EST (97 µg/m⁻³), ILH (59 µg/m⁻³), MOV (114 µg/m⁻³), and ERV (318 µg/m⁻³). At ERV, for instance, the measured daily average of PM₁₀ levels was approximately six times higher than the European air quality standard. This station is the closest to one of the EWEs. Regarding PM_2.5, only one air quality monitoring station was measuring this pollutant within the area affected by the EWE (i.e., EST) with a concentration of 52 µg/m⁻³ (

Table 3. The statistical performance indicators of the WRF-SFIRE system based on hourly results for the period between 15 and 16 October 2017 (24 h).

Pollutant	Station	r (-)	RMSE (µg/m3)	MB (µg/m3)	Average (µg/m3)
					WRF-SFIRE	Obs
PM10	EST	0.54	215	17	114	97
	AVE	0.30	300	52	160	108
	ILH	0.76	327	113	172	59
	ERV	0.35	418	−194	124	318
PM2.5	EST	0.68	219	59	111	52

).

The Portuguese air quality monitoring network recorded the daily average PM₁₀ levels exceeding the daily limit value (50 µg/m⁻³) set by the European Air Quality Directive (2008/50/EC) at several locations: EST (97 µg/m⁻³), ILH (59 µg/m⁻³), MOV (114 µg/m⁻³), and ERV (318 µg/m⁻³) (Table 3). At ERV, for instance, the measured daily average PM₁₀ levels were approximately six times higher than the European air quality standards. This station is the closest to one of the EWEs. Regarding PM_2.5, only one air quality monitoring station measured this pollutant within the area affected by the EWE (i.e., EST) with a concentration of 52 µg/m⁻³ (Table 3).

The WRF-SFIRE system simulated the highest daily PM₁₀ and PM_2.5 concentrations (>1000 µg/m³) in the regions affected by the fire progression (black areas delimited in Figure 2). Additionally, very high values were estimated over the northern area of Portugal, and the simulated values indicate that the central and northern regions of Portugal were also affected by PM₁₀ values larger than the air quality limit value established by the European Air Quality Directive for the protection of human health (50 µg/m³) (Figure 2). Despite the air quality monitoring network measuring the highest daily PM levels near the Portuguese coastline, the WRF-SFIRE simulations also predicted dangerous pollution levels inland where no monitoring stations were available.

To evaluate the performance of the WRF-SFIRE system in simulating the particulate matter emitted by the seven EWEs, Table 3 presents the hourly accuracy of the WRF-SFIRE system in simulating the PM₁₀ and PM_2.5 concentrations during the simulated period at the air quality monitoring stations affected by smoke (i.e., EST, AVE, ILH, and ERV).

The WRF-SFIRE system revealed reasonable accuracy in simulating the hourly PM₁₀ values, with a correlation coefficient ranging from 0.30 to 0.76 (AVE and ILH) and an RMSE varying between 215 µg/m³ and 418 µg/m³ (EST and ERV). For PM_2.5, these statistical parameters were 0.68 (r) and 219 µg/m³ (RMSE) for the only air quality monitoring station with measurements for this pollutant (EST). The modelling system tended to overestimate (positive MB) the PM levels, except for the values simulated for the ERV monitoring station, where the measurements (318 µg/m³) were higher than the simulated values (124 µg/m³).

4.2. Human Health

Considering the simulated levels of the PM_2.5 concentration, estimates were made for hospital admissions due to respiratory diseases and CVDs.

Table 4. Hospital admissions estimated in this study (15 and 16 October 2017) and reported by SNS (October 2017) by NUTSII and total.

NUTSII	Hospital Admissions (95% CI)
	Respiratory Diseases		CVDs (Including Stroke)
	This Study	SNS	This Study	SNS
Northern region	681 (95% CI: 0–1465)	2613	387 (95% CI: 72–709)	2970
Central region	766 (95% CI: 0–1691)	1558	429 (95% CI: 79–795)	1895
Alentejo	0 (95% CI: 0–0)	290	0 (95% CI: 0–0)	432
Total	1447 (95% CI: 0–3156)	4461	816 (95% CI: 151–1504)	5297

shows the estimated number of hospital admissions for respiratory diseases and CVDs (including stroke), by NUTS, caused by the PM_2.5 levels during the October 2017 EWEs in Portugal (this study). The number of hospital admissions registered by the Portuguese Health Service (SNS) in October 2017 is also presented. The results are presented for the most affected regions—north and central—and an area unaffected by smoke (Alentejo).

During the Portuguese EWEs (24 h period), a total of 1447 (95% CI: 0–3156) hospital admissions for respiratory diseases and 816 (95% CI: 151–1504) hospital admissions for CVDs (including stroke) were estimated. The northern region of Portugal accounted for about 47% of the total hospital admissions for respiratory diseases and CVDs, while the central region accounted for roughly 53%. On the other hand, no hospital admissions were estimated in Alentejo. The quantified total hospital admissions for respiratory diseases was about twice that of admissions for cardiovascular diseases (Table 4).

Figure 3 complements this information by showing the spatial distribution of hospital admissions related to the October 2017 EWEs.

The largest number of hospital admissions was estimated for the Leiria, Quiaios, and Lousã areas, where both the number of inhabitants (Figure A3) and the PM_2.5 concentration levels (Figure 2b) were high. As confirmed in the analysis of total hospital admissions, no hospital admissions were estimated in the Alentejo region. In contrast, the northern region was affected by smoke, resulting in estimated hospital admissions due to respiratory diseases and CVDs.

4.3. Visibility

Based on the estimated spatial and temporal levels of PM₁₀ and PM_2.5, visibility classes were calculated per hour and grid cell of the simulation domain. Figure 4 depicts the daily distribution, in percentage, of visibility classes for the total WRF-SFIRE grid cells of the simulation domain between 15 and 16 October 2017.

The poor visibility levels ranged from 1.3% (at 6:00 on 15 October 2017) to 33% (at 00:00 on 16 October 2017) for the WRF-SFIRE simulation cells (Figure 4). Visibility started to decrease at 12:00 on 15 October (5.1% of WRF-SFIRE cells with very bad visibility), with four out of seven active EWEs progressing due to favourable weather conditions. There were also increases in the number of firefighters (447 firefighters), aerial firefighting aircraft (maximum number of aircraft fighting the EWEs in the analysed period), and operational support materials (124 support materials) (Figure A4). Figure 5 shows the spatial distribution of the visibility classes at 12:00 on 15 October and 1:00 on 16 October.

At 12:00, the impact of the active fires on visibility was evident, with impaired areas (with poor and low visibility classes) extending downwind from the fire perimeter. At 1:00 on 16 October, the levels of visibility were the worst, with the entire northern region of Portugal and EWE areas having poor visibility levels (33% of WRF-SFIRE cells) (Figure 4 and Figure 5).

5. Discussion

This study highlights the significant impacts of EWEs on air quality, public health, and visibility, which can be estimated by the application of numerical modelling. Models are able to provide the spatial distribution of concentration levels of air pollutants and, therefore, of health and visibility impairment indicators. This is particularly important when there are not enough monitored data, as in this case study, where smoke from wildfires was affecting areas not covered by the Portuguese air quality monitoring network. Thus, monitored data did not fully capture the widespread impact of the EWEs. Moreover, equipment failure at the ERV monitoring station, where the PM concentrations reached the maximum measurable limit of 1000 µg/m³, highlights the need for policies that ensure the resilience of monitoring infrastructures under extreme air pollution conditions. This malfunction likely resulted from clogging due to heavy smoke, indicating the need for improved monitoring technologies or additional measures to address such failures in extreme smoke conditions.

Using remote sensing data for monitoring and evaluating model results is the way forward for exploration. This dual approach based on ground-based and remote sensing data could contribute to a more comprehensive validation of the model’s accuracy in depicting particulate matter distributions and their impacts.

Regarding the WRF-SFIRE modelling evaluation, and in spite of its reasonable performance, several factors may have contributed to numerical errors when comparing the model results with the measured values from the Portuguese monitoring network: (i) the lack of consideration of chemical reactions in smoke dispersion from the EWEs; (ii) the absence of atmospheric emissions from the main anthropogenic sources (e.g., residential, road transport, and industrial sectors) in the WRF-SFIRE simulations; (iii) the inaccurate quantification of wildfire emissions; (iv) errors in simulating the spread of wildfire and the dispersion of air pollution; and (iv) the application of the WRF-SFIRE system without a spin-up period due to computational constraints. Despite these identified modelling limitations, the system’s performance in this study aligns with previous air quality modelling applications for the simulation of PM dispersion during wildfire events. The daily and hourly modelling averages ranged from 12 µg/m³ to 44 µg/m³ [81] and from 7.16 µg/m³ to 812 µg/m³ [7,82], respectively, at the monitoring station locations, as reported in previous studies. Previous modelling efforts have also shown overestimation, with correlation coefficients ranging from 0.10 to 0.88 and RMSE values varying between 9.8 µg/m³ and 164 µg/m³ [7,12,57,81,82,83,84,85,86,87] (

Table A2. Numerical modelling performance in simulating PM₁₀ and PM_2.5 levels during wildfire events.

Pollutant	r (-)	RMSE (µg/m3)	MB (µg/m3)	Average (µg/m3)		Temporal Resolution	Model	Region	Reference
				Model	Obs
PM10	-	-	-	431–812	486	Hourly	DISPERFIRE	Central Portugal	[7]
PM10	-	11.10	−16.09	-	-	Hourly	LOTOS-EUROS + FLUENT	Coimbra, Portugal	[12]
PM10	0.78	32.22	−14.75	-	-	-	CHIMERE	Portugal	[83]
PM10	0.63	15.02	−30.35	-	-	-	LOTOS-EUROS	Portugal
PM10	0.88	70	-	-	-	Daily	CHIMERE	Russia	[84]
PM10	-	19.3	29.2	-	-	Hourly	LOTOS-EUROS	Portugal	[57]
PM10	-	17.0	20.7	-	-	Hourly	LOTOS-EUROS	Portugal	[57]
PM2.5	0.40	-	-	-	-	Daily	BlueSky	British Columbia, Canada	[85]
PM2.5	0.75	-	-	-	-	Daily	FireWork	British Columbia, Canada	[86]
PM2.5	0.58	-	-	-	-	Daily	BlueSky	British Columbia, Canada
PM2.5	0.16	-	−2.09	7.16	-	Hourly	RAQDPS	North America	[82]
PM2.5	0.41	-	−0.88	8.37	-	Hourly	FireWork	North America
PM2.5	0.19	11.90	−1.80	-	-	Hourly	ARQI	Washington State, USA	[87]
PM2.5	0.28	24.90	−1.18	-	-	Hourly	HRRR-Smoke
PM2.5	0.19	10.70	−2.34	-	-	Hourly	AIRPACT
PM2.5	0.10	42.90	3.38	-	-	Hourly	UCLA WRF-Chem
PM2.5	0.09	33.90	4.66	-	-	Hourly	UIOWA WRF-Chem
PM2.5	0.04	16.00	3.28	-	-	Hourly	WISC WRF-Chem
PM2.5	0.35	10.80	−0.32	-	-	Hourly	FireWork
PM2.5	0.26	10.40	0.02	-	-	Hourly	NAQFC
PM2.5	0.21	10.80	−5.78	-	-	Hourly	NCAR WRF-Chem
PM2.5	0.16	29.70	5.07	-	-	Hourly	GEOS-FP
PM2.5	0.29	12.50	5.51	-	-	Hourly	CAMS
PM2.5	0.11	9.80	−0.44	-	-	Hourly	RAQMS
PM2.5	0.15	32	−11.1	12	23	Daily	RAQDPS	Western Canada	[81]
PM2.5	0.57	101	21.1	44	23	Daily	FW-Ops
PM2.5	0.64	39	5.8	29	23	Daily	FW-CFFEPS
PM2.5	0.24	16	1.7	15	14	Daily	RAQDPS	Eastern Canada
PM2.5	0.26	16	2.3	16	14	Daily	FW-Ops
PM2.5	0.26	16	2.9	17	14	Daily	FW-CFFEPS
PM2.5	0.17	29	−8.0	15	23	Daily	RAQDPS	Western USA
PM2.5	0.49	164	21.1	44	23	Daily	FW-Ops
PM2.5	0.59	64	8.6	31	23	Daily	FW-CFFEPS
PM2.5	0.23	14	1.9	18	17	Daily	RAQDPS	Eastern USA
PM2.5	0.23	14	2.1	19	17	Daily	FW-Ops
PM2.5	0.24	14	2.8	19	17	Daily	FW-CFFEPS

). These findings suggest that, despite the need for further modelling refinements to enhance the accuracy of simulations, the results demonstrate that such tool can be valuable for pollutant dispersion forecasting and for informing mitigation strategies during EWE. Integrating these tools into decision-making processes and policy planning could enhance wildfire response, particularly in protecting public health and managing visibility.

The estimated hospital admissions during the 24 h EWE period were lower (1447 and 816 hospital admissions) than those reported by the National Health Service (SNS) in October 2017 (a one-month period), which recorded 4461 and 5297 hospital admissions for respiratory and cardiovascular diseases, respectively. Despite the different time periods, the estimated hospital admissions for respiratory diseases and CVDs (including stroke) were around 34% and 20%, respectively, of the total hospital admissions registered by the SNS in October (Table 4). This comparison also confirms the expected coherence between the estimated and reported data, with estimations for a 24 h period being below the monthly reported values but still in the same order of magnitude. No hospital admissions were estimated in the Alentejo region, as the smoke from the EWEs was transported northward by strong winds from Hurricane Ophelia (Figure 2b) [35].

In addition to morbidity impacts, several studies have estimated the mortality effects of PM exposure. For example, Arregocés et al. [88] quantified 738 deaths per year directly attributed to PM exposure in the northern Caribbean region of Colombia, while Zhao et al. [89] found that, at a global level, 2.97 million deaths were associated with PM_2.5 exposure as of 2018. Similarly, Gouveia et al. [90] assessed the short-term effects of PM_2.5 exposure in 337 cities across nine Latin American countries, estimating 3,026,861 cardiovascular deaths and 1,222,623 respiratory deaths. These findings highlight the substantial health burden posed by PM, both in the short and long term, reinforcing the importance of addressing air quality to mitigate both acute and chronic health risks. Despite some uncertainty, the public health impacts, as evidenced by the estimated hospital admissions during the 24 h EWE period, reinforce the need for public health policies that incorporate smoke dispersion forecasts and integrate these data into emergency planning. Such policies could mitigate adverse health effects, particularly in vulnerable populations. The obtained visibility results were also consistent with those of previous studies. Valente et al. [7] applied a similar approach to quantify the visibility impairment caused by an experimental fire and obtained values ranging from 24 dV (i.e., clear visibility) to 70 dV (i.e., poor visibility). During the night, atmospheric stability caused smoke to accumulate at lower altitudes (<2000 m), reducing visibility and potentially hindering firefighting efforts. Although aircraft could not be deployed for nighttime operations, the highest number of firefighters and support resources were mobilised during this period (Figure A4), highlighting the intense response required to manage the EWEs’ impacts effectively. These visibility results underscore the importance of developing policies that improve communication regarding visibility conditions to the public and emergency response teams. Implementing more accurate alert systems that account for extreme conditions during EWEs could aid in managing air and ground traffic and enhance the safety of firefighting operations.

This study presents an updated method to evaluate the impact of EWEs on human health and visibility. However, in addition to the limitations already mentioned regarding the WRF-SFIRE system’s performance, some others had to be assumed, such as the following: (i) double-counting issues could arise since the population census dataset already includes the firefighters’ residences; (ii) the geographical location of each firefighter was unknown, so the hourly geographical locations of the fire occurrences were used as proxies for the temporal and spatial distributions of the firefighter teams; (iii) no human time–activity patterns were considered, meaning that air pollution levels could change in space and time during the population’s daily activities/routines, influencing their exposure; (iv) to assess human health impacts, outdoor air pollution concentrations from the WRF-SFIRE system were considered, although people typically spend most of their time indoors; (v) concentration–response functions and relative risks were not derived for this type of extreme wildfire event, which implies that there were very high concentrations of pollutants during a short period of time; (vi) health impacts were estimated by considering only PM_2.5, even though the health impacts of smoke result from exposure to several pollutants that can have synergistic effects; (vii) the evaluation of health results was constrained by the available monthly health data, which did not allow for a direct comparison with daily modelling results during the EWEs; and (viii) visibility impairment estimates were based on PM₁₀ and PM_2.5 only, and the equations used were not specifically derived for highly polluted areas. This list of limitations did not strongly affect the quality of the results and the main outcomes of this study. They should mainly be considered as opportunities and suggestions to improve further studies and to obtain even more solid results on the impacts of smoke from wildfires on human health and visibility.

In summary, the findings of this study indicate that improvements in environmental and public health policies are necessary to address the challenges posed by EWEs. The development of integrated policies that combine enhanced monitoring infrastructures, the use of advanced predictive models, and the implementation of more robust alert and response systems is recommended. Additionally, further research on the relationship between smoke concentration levels of several pollutants (e.g., PM₁₀, PM_2.5, NO₂, or even polycyclic aromatic hydrocarbons) and health response functions would be valuable, as well as the development of visibility impairment equations adapted to very polluted areas.

6. Conclusions

The main goals of this work were to quantify, with a high spatial resolution, the atmospheric pollution caused by an EWE and assess its short-term impacts on human health (population and firefighters) and visibility. The Portuguese EWEs in October 2017 were selected as the case study, and the WRF-SFIRE system was used to simulate smoke dispersion, while the human health impacts and visibility impairment were estimated using up-to-date numerical approaches.

The WRF-SFIRE system simulated the highest daily PM₁₀ and PM_2.5 concentrations (>1000 µg/m³) in the regions affected by the fire progression, but also estimated dangerous levels (>50 µg/m³) in the central and northern regions of Portugal. The air quality modelling performance was consistent with previous studies. The system tended to overestimate the PM levels but demonstrated reasonable accuracy in simulating pollution dispersion (r ≥ 0.30 and RMSE < 420 µg/m³). The modelling errors could be attributed to the assumptions that the smoke emitted does not react chemically in the atmosphere, the lack of consideration of atmospheric emissions from the main Portuguese anthropogenic sources in the numerical simulation, the inaccurate quantification of wildfire emissions, and errors in simulating both the spread of wildfire and the dispersion of air pollution.

Based on the PM simulation results, a total of 1447 hospital admissions for respiratory diseases and 816 hospital admissions for CVD (including stroke) were estimated, with the central region of Portugal accounting for about 53% of the total admissions for both health endpoints analysed. These numbers represent 32% and 15% of the total reported SNS admissions in the region during October.

The simulated PM values also allowed us to estimate visibility impairment during the EWEs. Visibility started to decrease at midday (with four out of seven EWEs active), and the worst level of visibility was reached at night, with the entire northern region of Portugal and EWE areas registering poor visibility levels (with a visual range of less than 2 km). Although aircraft cannot be operated to fight wildfires during the night, the highest numbers of firefighters and support materials were recorded during this period. Even though the methodology used was not designed for these specific smoke events and further investigation is required to better represent the levels of air pollutant concentration that affect visibility, simulating and forecasting visibility impairment over time and space could provide valuable support for firefighting operations and civil protection efforts.

This work highlights the potential of numerical models to predict, with high spatial and temporal resolutions, the population potentially exposed to critical levels of air pollution due to active wildfires as well as to provide useful information to citizens, civil protection agencies, and health entities to drastically reduce the impact of wildfires on society.