The Role of Field Measurements of Fine Dead Fuel Moisture Content in the Canadian Fire Weather Index System—A Study Case in the Central Region of Portugal

Abstract

The Canadian Fire Weather Index System (CFWIS), empirically developed for forests in Canada, estimates the fuel moisture content ($m_f$) at different depths and loads through meteorological parameters. While it is often suggested that adapting an existing fire danger rating system like CFWIS for a new environment requires developing new relationships or modifying existing ones, it is worth considering if such adaptations are always necessary. Based on a dataset of field measurements for surface litter (Pinus pinaster) carried out in the central region of Portugal (2014–2023), we propose a correction of $m_f$ based on the Fine Fuel Moisture Code (FFMC) of the CFWIS. This moisture correction was used to determine the Initial Spread Index (ISI) directly and, subsequently, the Fire Weather Index (FWI). Fire records from the study region were used to analyze the performance of the corrected indices. We found that the moisture correction led to higher values and potentially more accurate indices under dry conditions but did not provide a significant improvement in predicting the number of fires and burned areas compared to the original indices. The results suggest that, in relation to fire activity, the CFWIS is sufficiently robust to variations in the fuel moisture content in the study region.

1. Introduction

1.1. Background

The increase in global temperatures and reduction in precipitation suggest that meteorological conditions conducive to wildfires are becoming more frequent [1]. Climate change has led to more frequent, longer, and more intense heat waves and droughts, increasing fire danger with longer fire seasons, more frequent large wildfires, more severe fire behavior episodes, and new fire-prone regions [2]. This change is expected to increase fire danger across most of Europe, especially in southern regions where fires are already frequent and intense [3]. Thus, it is crucial to analyze the fire danger systems in the context of climate change.

Meteorological parameters are used as inputs in fire danger systems to indicate the physical probability that a fire will start and propagate, known as fire danger [4]. Systems such as the United States National Fire Danger Rating System (first introduced in 1964, [5]), Australian McArthur Forest Fire Danger Index (1967, [6]), and Canadian Fire Weather Index System (1987, [7]), typically estimate the weather effect on the moisture content of live and dead vegetation. This vegetation is considered as “fuel”, which includes any organic material—live or dead: (i) on the ground (duff, organic soil), (ii) on the surface (litter, shrubs), or (iii) in the air (crown fuels). Fuel moisture content ($m_f$) represents the amount of water per dry mass of the fuel, expressed as a percentage, and directly influences the ease of ignition and the ability to sustain the fire propagation. The $m_f$ is not in itself a meteorological factor, but it is a product of the cumulative effects of past and present meteorological events [8]. In terms of fire behavior assessment, live and dead fuel moisture must be distinguished since they have different water retention mechanisms and different responses to meteorological conditions [8,9,10].

Live fuels have moisture contents that result from the interaction of physical and physiological processes [11], and their flammability ranges from about 35% to well over 200% [8]. The moisture content of live fuels is not commonly included in fire danger systems, as it is more difficult to estimate from meteorological data compared to dead fuel moisture [9]. However, the foliar moisture content (live fuels) is one of the factors influencing the onset or initiation of crowning in coniferous and mixed-wood fuel types [12]. Although it is beyond the scope of this study, [13] recommendations for obtaining the foliar moisture content based on a review of the pertinent literature are provided.

Fine dead fuels are characterized by the vegetation materials lying on the forest litter with a diameter of less than 0.6 cm, such as needles, leaves, and herbaceous plants. According to the definition adopted for the U.S. National Fire Danger Rating System [5], fine dead fuels are often considered to have an inherent response time of one hour [5], as these fuels respond more quickly with precipitation and other changes in the atmospheric moisture content, taking 1 hour to adjust to moist/dry conditions. Those fuels change their water content in parallel to atmospheric conditions [9,10]. Consequently, fine dead fuels are drier and more prone to ignition than live fuels [9,10]. Dead forest fuels, mainly fine fuels, are responsible for sustaining wildfires and determining their rate of spread [14]. Litter and small fuels ignite and burn easier and faster than large fuels as less heat is required to remove fuel moisture and raise the ignition temperature. The fine dead fuel moisture content range is about 1.5 to 30 percent [8], and values below 5% indicate a high probability of having extreme fire behavior (crown fires, spotting, and very high rates of spread) [15,16].

The dead fuel moisture content is one of the most used and oldest indicators in predicting fire danger [17] and is a key parameter in assessing the potential for large wildfires. The moisture content of dead fuels can be measured in the field (e.g., [18,19,20]), estimated using fuel moisture indicator sticks (e.g., [21]), or estimated indirectly by various models ([22,23]). However, obtaining accurate results for the evolution of the fine fuel moisture content remains a complex task due to its interaction with meteorological, topographic, and vegetation factors [24]. Fire danger systems commonly ignore fine-scale, topographically induced weather variations [25]. This is acceptable when the purpose is to estimate the characteristic $m_f$ value for a geographical area rather than a specific place. Prediction models usually have a small number of inputs, considering the several parameters that affect the moisture content of dead fine fuels [18]. Using inputs measured in the field is a necessary step to have a better understanding of fuel moisture content and to assess the applicability of modeling tools [23].

The present study aims to understand the role of field measurements of fine dead fuel moisture content in the Canadian Fire Weather Index System (CFWIS), a fire danger system that was empirically developed for boreal forests in Canada but is used in many parts of the World. One of the objectives of this study is to assess the robustness of the original calibration performed during the development of the CFWIS using data obtained in a specific environmental condition.

1.2. Canadian Fire Weather Index System

The CFWIS, developed by the Canadian Forestry Service in 1970 [26], is a fire danger system that results from decades of research about the fundamental relationships between meteorological parameters and fuel moisture [27]. Based on the 1987 version [7], the CFWIS uses meteorological data as input parameters to estimate the moisture content of dead fuels in different soil layers. The meteorological data (temperature, relative humidity, precipitation, and wind) are measured daily at 12:00 (local standard time) at weather stations that are intended to represent the average conditions in a certain region.

The primary outputs of the CFWIS are the fuel moisture codes, each with different drying rates, nominal fuel depth, and nominal fuel loads: the FFMC—Fine Fuel Moisture Code, DMC—Duff Moisture Code, and DC—Drought Code. These codes are numerical ratings of the fuel moisture content of fine surface litter (FFMC), loosely compacted duff of moderate depth (DMC), and deep compact organic matter (DC) [27]. Each code is calculated daily and includes the previous day’s value as an input to the current day’s value [6]. They also consider the duration of the day from sunrise to sunset. Because they are cumulative, taking into account the previous day’s value, two or three good days of drying following heavy rain would produce a high FFMC while the DMC remains low [27]. Conversely, a light rain after a long dry spell will result in a low FFMC while the DMC remains high [27]. The DC may rise or fall slowly, while the FFMC and DMC fluctuate more frequently [27]. These codes are different indicators of fire potential: FFMC indicates the relative ease of ignition and the incidence of spot fires [28]; DMC typically indicates the ease of ignition by lightning [29]; and DC indicates long-term drought conditions, used to estimate when a particularly severe fire season is imminent [30,31].

The intermediate outputs of the CFWIS are the fire behavior indices: the Initial Spread Index (ISI), which is a combination of the effects of wind speed and the fine dead fuel moisture content on fire spread, and Build Up Index (BUI), which is a combination of the DMC and DC that represents the fraction of the total fuel that is available to the fire spread. ISI and BUI are then combined to provide the final output parameter of the system: the Fire Weather Index (FWI). The FWI is a fire behavior index that represents Byram’s (1959) fire intensity, defined as the energy output per unit of time and length of the fire front [32]. Since fine dead fuel moisture content in the CFWIS is directly considered in the FFMC and ISI and subsequently in the FWI, we only present the final equations of these components below to maintain the conciseness of the article. These equations result from a set of calculations with specific conditions for application that are described in [7].

The current standard FFMC was derived from the original work by [33,34], who produced their first set of fire danger tables relating the moisture content of needle litter and top-layer duff in red and white pine stands. The FFMC represents the moisture content of surface litter and other fine fuels in a layer with a dry weight of around 0.25 kg·m⁻² [7]. The rising relevance of shorter-term water balance in the development of severe fire can be observed in the FFMC, which is the fastest-drying component of the CFWIS [17]. Meteorological parameters are used in a set of equations presented in [7] to determine the fine dead fuel moisture content based on meteorological data ($m_{f_{FFMC}}$) and thus the FFMC value. FFMC (as well as DMC and DC) is arranged such that it increases in value as moisture content decreases. This inverted scale was developed for psychological effect, so that higher numbers indicate higher fire potential [35].

The first of the two intermediate components is the Initial Spread Index (ISI), which combines the effects of wind speed and fine fuel moisture content on fire spread according to the equations presented in [7].

The final output, the FWI, is determined from ISI and BUI according to the equations presented in [7]. The FWI is a numerical rating that represents Byram’s (1959) frontal fire intensity [7]. The FWI is a relative measure of the potential intensity of a single spreading fire in a standard fuel complex on level terrain [27] based on meteorological data. An increase in the FWI (or any other index of CFWIS) corresponds to an increase in fire danger, and it is usually classified according to five or six classes: Low, Moderate, High, Very High, Extreme, and Very Extreme [36].

The FWI is the component of CFWIS commonly used as a measure of general fire danger for operational purposes, as it is a good indicator of several aspects of fire activity. However, the remaining components of the system also need to be examined, as they provide direct information about certain aspects of wildfire potential [27].

Modifications in Moisture Codes of the CFWIS

The moisture codes of the CFWIS were developed to be representative of the moisture content of forest floor fuels in a mature, closed-canopy jack pine (Pinus banksiana Lamb.) or lodgepole pine (Pinus contorta Dougl.) stand [35]. CFWIS is adaptable enough to have been implemented outside of forests in Canada and in countries with very different climates [6]. However, this implementation usually requires modification for the region where it is applied [37]. These modifications are made through new relationships between the moisture code and the moisture content of the fuel layer in a specific region to indicate fire potential in the dominant forest fuel type of the area, especially when that forest floor is significantly different from the standard CFWIS closed-canopy pine stand [35].

Both in Canada and other countries, new relationships are often established between one or more of the six components of the CFWIS [31]. Some of the modifications to the FFMC are worth mentioning here:

• A function for estimating the moisture content of fine dead fuels from the FFMC was defined by [38]. The results were obtained through field measurements of Pinus pinaster and Eucalyptus globulus samples collected in Lousã (central region of Portugal);
• An indicator of grass ignition potential in Sumatra (Indonesia) using the FFMC was developed by [39];
• In Canada, [35] presented the development of explicit relationships between observed litter moisture and the FFMC for several major forest types across the country (jack pine, spruce, Douglas fir, mixed-wood, and deciduous). The proposed models for litter moisture can be used in these types of stands when actual litter moisture estimates are required for specific fire management or fire research applications;
• The assessment of the efficiency of using the FFMC in an equatorial climate area and the development of a new model for estimating the dead fine fuel moisture content using fuel samples was conducted by [40].

While it is often suggested to adapt an existing fire danger rating system to a new environment by considering new relationships or modifying existing ones as necessary [41], it is worth considering whether such adaptations are always required. This raises the question of whether existing systems like CFWIS might already have sufficient robustness to handle variations in moisture content without extensive modifications.

1.3. Objectives

CFWIS estimates the moisture content of fine dead fuels based on meteorological parameters in the FFMC routine. Despite being widely used outside of Canada, this parameter is generally overestimated in the FFMC routine [35,40], leading to a lower danger level than actually exists. Using a dataset of fuel moisture content from an important litter component in the central region of Portugal (dead needles of Pinus pinaster) collected between 2014 and 2023 (10 years), we intend to compare those measurements with the moisture content of dead fine fuels estimated by the CFWIS. Since there are significant differences between these variables (measured vs. estimated), we developed a similar equation proposed by [38] using recent field data to correct the estimated fuel moisture based on the FFMC. This moisture correction is then directly used in the calculation of the ISI and, consequently, the FWI. These indices using moisture correction are compared with the original indices of the CFWIS. Finally, we compare the performance of the corrected values of ISI and FWI with fire activity in terms of the number of fires and burned area.

2. Materials and Methods

2.1. Study Region

The study takes place in the central region of Portugal, where the climate is Mediterranean, characterized by long dry summers and rainy winters that promote vegetation growth. Pinus pinaster and Eucalyptus globulus are the predominant species, mostly in mixed stands [23,42]. The fire regime has varied and is associated with climate change, with long drought periods and frequent heat waves increasing fire occurrence and severity. Historically, the fire season in mainland Portugal typically spans from July to September. However, it now begins in June and ends in October [43], as evidenced by the large wildfires that occurred in June and October of 2017.

In administrative terms, the central region of Portugal covers 100 municipalities in the districts of Coimbra and Castelo Branco and most of the districts of Aveiro, Viseu, and Guarda [44]. The study region has a total area of 201,000 ha (2010 km²) and comprises eight municipalities of the Coimbra district (Lousã, Miranda do Corvo, Penacova, Góis, Penela, Vila Nova de Poiares, Arganil, and Coimbra). This region is located between latitudes 39°55′ and 40°25′, and longitudes 8°35′ and −7°45′ (Figure 1).

2.2. Dataset and Statistical Analysis

The following datasets covering the period from 2014 to 2023 are used:

(i)

Meteorological data measured at the weather station of Lousã (ID: 697) by the Portuguese Institute of the Sea and Atmosphere (IPMA);

(ii)

Field measurements of moisture content of dead needles (or leaves) of Pinus pinaster ($m_{f_{PP}})$ by the Forest Fire Research Center of Association for the Development of Industrial Aerodynamics (CEIF-ADAI) [19,45];

(iii)

Fire records (number of fires and burned area) for the study region by the Institute for the Conservation of Nature and Forests (ICNF), the entity of the Portuguese Government responsible for reporting the official statistics of wildfires [46].

According to these datasets, Figure 1 presents the location of the sampling plot in Lousã where field measurements were made, the weather station used (around 600 m distance from the sampling plot), and the municipalities that compose the study region where fire records are considered.

The statistical analysis of the data was conducted using IBM SPSS Statistics software (Version 28). Statistical tests to analyze the distribution of independent groups (e.g., estimated moisture content of dead PP by meteorology vs. measured moisture content of dead PP in the field) were performed using the Mann–Whitney U test, with a significance level (α) set at 0.05. The Mann–Whitney U test, a non-parametric alternative to the Student’s t-test, was used to compare the significance of differences between groups. In contrast to Student’s t-test, which is a parametric test for comparing means, the Mann–Whitney U test is based on ranks, does not assume a normal distribution of the data, and does not assume the homogeneity of variances. The Mann–Whitney U test is more suitable for our analysis since the fuel moisture data are not normally distributed and contain some strong outliers. In this case, it is more appropriate to use ranks rather than actual values to avoid the testing being affected by the presence of outliers or by the non-normal distribution of data.

2.2.1. Meteorological Data

Meteorological data from the weather station in Lousã (ID: 697), located approximately 600 m from the sampling plot, were provided by IPMA, the national authority for weather and climate, and manages a wide network of ground sensors [47]. Daily meteorological data (air temperature, air relative humidity, wind speed, and precipitation) measured at 12:00 (local standard time) were used to determine the indices of the CFWIS. A summary of meteorological conditions (Figure A1) for the study period can be found in Appendix A.

For the total area of the study region (2010 km²), only the weather station in Lousã was used. In [48], a similar approach was followed, where the authors used one weather station per Portuguese district (e.g., Coimbra district has 3947 km²) to analyze the CFWIS in relation to fire activity (number of fires and burned area). Also, according to [49], it was found that the data of a single weather station are representative of a wide area in the central region of Portugal.

2.2.2. Field Measurements of Fine Dead Fuels Moisture Content

Pinus pinaster (PP) and Eucalyptus globulus (EG) are very common species in forests of Central Portugal, as well as in other areas of the Mediterranean basin, and both litters are directly involved in the ignition and propagation of wildfires [18,23]. Their leaves are considered fine fuels, and according to [5], the dead leaves are classified as 1-h fuels. Pine needle litter forms a highly porous fuel bed and provides a source of continuous fuel cover [50], while eucalypt litter is not homogeneous and can be stratified into a relatively compacted horizontal surface fuel layer [51]. The present study focused exclusively on Pinus pinaster litter, as no difference was found in practical terms in the distribution of $m_f$ values obtained with Pinus pinaster or Eucalyptus globulus [38].

The sampling process used in this study is a direct method that involves the collection of dead needles of PP to obtain their fuel moisture content. Samples are collected in a plot in Lousã, consisting of a stand with a mixture of Pinus pinaster and Eucalyptus globulus, with an average tree height of about 10 m and a crown closure of around 35% [18]. The shading on the ground was not changed during the period of the study. Various shrubs (live fuels) in the plot are also sampled. Additional information about vegetation mapping in Lousã can be found in [42], and a detailed characterization of the forest litter in the central region of Portugal can be found in [23,52]. The plot has a southern exposure and slope of 30%, representing one of the worst scenarios of litter dryness, which is usually not considered in the prediction models based on meteorological data. In [25], the influence of topographic variation on fuel moisture is analyzed.

Samples were collected between 12:00 and 14:00 (local standard time) every day during the main fire season (15th of May–15th of October) and twice a week for the rest of the year. Even if there was a precipitation event on the same day or days before, fine dead fuels are collected as they dry very fast [53]. Some daily gaps in the $m_f$ measurements occurred since the collection was primarily carried out on working days. After being collected, samples were transported in an isothermal bag immediately to the Forest Fire Research Laboratory (LEIF). At LEIF, four samples of dead needles of PP, each weighing 5 g (initial mass: $m_i$), were dried in an oven for at least 24 h at 105 °C. After this time, their dry mass ($m_d$) was weighed.

The moisture content of dead needles of Pinus pinaster, $m_{f_{PP}}$, in percentage for each sample is determined through the gravimetric method (Equation (1)). The daily value of $m_{f_{PP}}$ results from the average moisture values of the four samples. During the study period (2014–2023), 1018 measurements of $m_{f_{PP}}$, were made, corresponding to the number of days on which sampling was physically carried out.

$$m_{f_{PP}}={\displaystyle \frac{m_i-m_d}{m_d}}\times 100 \left[\%\right].$$

(1)

Figure 2 presents (a) the stand complex of Pinus pinaster and Eucalyptus globulus at the sampling plot, (b) the dead forest litter, and (c) 5 g samples of dead Pinus pinaster needles prepared at LEIF, which is located less than 1 km from the sampling plot.

The fine dead fuel moisture content estimated by the meteorological data, $m_{f_{FFMC}}$, and the field measurements of moisture content of dead needles of Pinus pinaster, $m_{f_{PP}},$ were compared. For operational purposes, fire managers are most interested in fuel moisture at the dry end of the scale as it may affect the ignition potential or rate of spread of a large fire [35]. Similar to the approach in [35], to ensure that our data are most accurate at this dry end, in the development of the correction, we only considered the days when $m_f$ values were lower than 28.50%, which is equivalent to a FFMC higher than 75 in the standard CFWIS. This condition resulted in a sample size of 812 days in the group of $m_{f_{FFMC}}$ values and a sample size of 839 days in the group of $m_{f_{PP}}$ values. Considering the 1018 days, there were more days with values lower than 28.50% in the $m_{f_{PP}}$ group than in the estimated $m_{f_{FFMC}}$.

Figure 3 presents the histogram, accompanied by the boxplot, illustrating the moisture content of (a) $m_{f_{FFMC}}$ and (b) $m_{f_{PP}},$ both conducted between 2014 and 2023. The histogram binning (number of classes) was determined using the Sturges’ rule [54]. The boxplot provides complementary information by depicting the central tendency and potential outliers within the dataset. The descriptive statistics for each group (mean and median, among others) are presented in Appendix A in

Table A1. Descriptive statistics of values of fuel moisture content ($m_f$) lower than 28.50% for samples of Pinus pinaster ($m_{f_{PP}}$) and $m_f$ estimated by the FFMC ($m_{f_{FFMC}}$).

${\mathit{m}}_{\mathit{f}}$			Statistic	Std. Error
${\mathit{m}}_{{\mathit{f}}_{\mathit{P}\mathit{P}}}$ (n = 839)	Mean		12.286	0.182
	95% Confidence Interval for Mean	Lower Bound	11.929	-
		Upper Bound	12.643	-
	Median		10.990	-
	Variance		27.771	-
	Std. Deviation		5.270	-
	Minimum		3.682	-
	Maximum		28.490	-
	Range		24.808	-
	Interquartile Range		5.964	-
	Skewness		1.132	0.084
${\mathit{m}}_{{\mathit{f}}_{\mathit{F}\mathit{F}\mathit{M}\mathit{C}}}$ (n = 812)	Mean		12.779	0.168
	95% Confidence Interval for Mean	Lower Bound	12.449	-
		Upper Bound	13.110	-
	Median		12.111	-
	Variance		22.993	-
	Std. Deviation		4.795	-
	Minimum		2.764	-
	Maximum		28.433	-
	Range		25.669	-
	Interquartile Range		5.653	-
	Skewness		0.811	0.086

.

The normality of the $m_{f_{PP}}$ data and $m_{f_{FFMC}}$ data (both with p-value < 0.001) and homogeneity of variances (p-value = 0.017) across comparison groups were not verified when the statistical tests were performed with a significance level of 0.05. The outputs of the statistical test are presented in Appendix A (

Table A2. Output of the statistical test for the (a) normality of the variables ($m_{f_{PP}}$ and $m_{f_{FFMC}}$) based on Kolmogorov–Smirnov and (b) homogeneity of variances based on Levene’s statistic.

(a)
Tests of Normality
Kolmogorov–Smirnov a
${\mathit{m}}_{\mathit{f}}$	Statistic	df		p-Value
$m_{f_{PP}}$	0.126	839		<0.001
$m_{f_{FFMC}}$	0.081	812		<0.001
(b)
Tests of Homogeneity of Variances
Levene
${\mathit{m}}_{\mathit{f}}$	Statistic	df1	df2	p-Value
Based on mean	5.742	1	1649	0.017
Based on median	2.550	1	1649	0.110

^a Lilliefors significance correction.

). Given these conditions, we used the non-parametric Mann–Whitney U test to explore significant differences between the groups ($m_{f_{PP}}$ and $m_{f_{FFMC}}$).

The results of the Mann–Whitney U test (

Table A3. Output of the Mann–Whitney U test between $m_{f_{PP}}$ (n = 839) and $m_{f_{FFMC}}$ (n = 812).

Independent-Samples Mann–Whitney U. Test Summary
Total N	1651
Mann–Whitney U	377,715
Wilcoxon W	707,793
Test Statistic	377,715
Standard Error	9684.415
Standardized Test Statistic	3.829
p-value Asymptotic Sig. (2-sided test)	<0.001

) indicate that there is a statistically significant difference (p-value < 0.001) between the groups in relation to the moisture content variable. The moisture values estimated by the FFMC tend to have higher ranks than the values measured in the field, meaning that the estimated values are significantly higher than the moisture values observed for PP. Considering this significant difference, we developed a correction to the fine fuel moisture content ($m_{f_{Corr}}$) based on the relationship between the $m_{f_{PP}}$ and the FFMC.

2.2.3. Fire Records

Eight municipalities comprise the study region for the fire records analysis, selected due to the necessity of obtaining a sufficient number of fires (NF) and burned area (BA) for a robust statistical analysis. The dataset of fire records provided by ICNF [46] includes detailed information on fire ignitions, such as the day and time of ignition, the location at different levels (e.g., the municipality where it occurred), and the burned area associated with each ignition.

For each municipality shown in Figure 1, the NF and BA in hectares were summed on a daily basis. From here onwards, fire records are reported for the entire study region without distinctions between municipalities. Our analysis revealed that 98% of the fire occurrences and 99% of the burned area took place on days when the fuel moisture content ($m_f$) was below 28.50%.

Appendix A contains the daily number of fires (Figure A2) and the daily burned area in hectares (Figure A3) for the study region.

2.3. Methodology

2.3.1. Development of Fine Fuel Moisture Content Correction ($m_{f_{Corr}}$)

The meteorological data registered in Lousã weather station (ID: 697) are used to determine the original indices of the CFWIS [7] for all days (n = 3652). The correspondence between the FFMC and the $m_{f_{PP}}$ for the same day resulted in a dataset with 1018 days, which are the days for which we have field measurements. The days with $m_{f_{FFMC}}$ and $m_{f_{PP}}$ lower than 28.50% were selected, resulting in 744 days for analysis.

The methodology to determine fine fuel moisture content correction follows a percentile-based approach adapted from [55,56], initially developed by [38] to calibrate the FWI for the Portuguese districts, using the following steps:

i.

Ordering fuel moisture data by descending order and adding the probability variable

The data are ordered by decreasing values of $m_{f_{PP}}$. Afterward, we add two new variables: the numerical incremental day and the probability. The numerical incremental day ($d_n$) value of “1” is attributed to the record with the highest value of $m_{f_{PP}}$, and consecutively adding “1” to the following record; the last record corresponds to the total number of days ($d_{Total}=744)$. The probability (P) is determined according to Equation (2) and reflects the weight that a given day has with respect to the total number of days.

$$P={\displaystyle \frac{d_n}{d_{Total}}}.$$

(2)

The lowest $m_{f_{PP}}$ corresponds to the day $d_{744}$ with the highest respective value of probability. Consequently, the fire danger is greater on the days with the highest probability values.

ii.

Grouping the variables by probability classes

Grouping the variables ($m_{f_{PP}}$ and FFMC) into probability classes is equivalent to splitting the results by percentiles. In [38,56], the probability values were arranged into 11 classes (P10, P20, P30, P40, P50, P60, P70, P80, P90, P95, P > 95). This arrangement results in only 5% of the days for the last two percentiles (P95, P > 95) since, according to [38], the days of maximum danger should be fewer to avoid distorting their meaning. To establish a relationship between $m_{f_{PP}}$ and FFMC, we calculated the average values of the variables for 25 classes and 40 classes, each with an approximately equal number of days, in addition to the usual arrangement of 11 classes [38,56]. Similar to a histogram, increasing the number of classes allows for a better observation of the distribution of $m_{f_{PP}}$ values.

The probability values divided into 25 classes (or 25 percentiles) with an amplitude of probability equal to 0.04 are presented in Figure 4. This figure shows the mean value of fuel moisture content of Pinus pinaster ($m_{f_{PP}}$) for each percentile group, with the bars representing the standard error of the mean. In Appendix A,

Table A4. Data arranged into 25 probability classes: number of days per class (n) for a total number of days ($d_{Total}$) of 774, mean values of observed moisture content of dead Pinus pinaster ($m_{f_{PP}}$) and mean values of the Fine Fuel Moisture Code (FFMC).

Probability Class	Percentile	n	$\mathbf{Mean} \mathbf{of} {\mathit{m}}_{{\mathit{f}}_{\mathit{P}\mathit{P}}}$	Mean of FFMC
[0.00–0.04]	P4	30	25.99	82.96
]0.04–0.08]	P8	31	21.60	82.97
]0.08–0.12]	P12	31	18.48	84.67
]0.12–0.16]	P16	31	16.34	85.83
]0.16–0.20]	P20	31	15.11	86.35
]0.20–0.24]	P24	31	14.16	87.14
]0.24–0.28]	P28	31	13.40	88.52
]0.28–0.32]	P32	31	12.72	88.24
]0.32–0.36]	P36	31	12.25	87.82
]0.36–0.40]	P40	31	11.83	87.09
]0.40–0.44]	P44	31	11.41	88.87
]0.44–0.48]	P48	31	10.99	89.41
]0.48–0.52]	P52	31	10.53	89.62
]0.52–0.56]	P56	31	10.13	88.65
]0.56–0.60]	P60	31	9.82	88.91
]0.60–0.64]	P64	31	9.52	89.70
]0.64–0.68]	P68	31	9.11	90.13
]0.68–0.72]	P72	31	8.72	89.43
]0.72–0.76]	P76	31	8.40	90.56
]0.76–0.80]	P80	31	8.10	91.09
]0.80–0.84]	P84	31	7.78	91.20
]0.84–0.88]	P88	31	7.37	91.38
]0.88–0.92]	P92	31	6.86	91.64
]0.92–0.96]	P96	31	6.23	92.11
]0.96–1.00]	P > 96	31	5.06	93.59
		$d_{Total}=774$

presents the data arranged into 25 probability classes where each $m_{f_{PP}}$ average value has a given probability of occurrence that is indicated by the percentile value. For example, a percentile P60 refers to a $m_{f_{PP}}$ equal to 9.82%, which has a probability of occurrence equal to or lower than 0.60.

We observe a decreasing trend in the mean values of $m_{f_{PP}}$, which is expected due to the data being ordered. Since the number of observations is constant in each group, the decrease in the amplitude of the error bars (which becomes less visible from the P16 onward) with increasing percentiles indicates that there is less variability in $m_{f_{PP}}$ values in the higher percentiles, resulting in more precise means for drier conditions. Overall, there is little variability in values within each percentile group, which demonstrates the reliability of this approach for analyzing the data. Based on this percentile approach, the function $m_{f_{PP}}=f_1\left(FFMC\right)$ was applied where a $m_{f_{Corr}}$ lower than 28.50% corresponds to an FFMC higher than 82. The fine fuel moisture content correction is presented in the dedicated Section 3.1.

2.3.2. Determination of the Indices Using the Fine Fuel Moisture Correction

During 2014–2023, we determined the indices using the proposed fine dead fuel moisture correction for as many days as possible to analyze them in relation to fire activity. On days with and FFMC higher than 82 (n = 1993), the corrected moisture content ($m_{f_{Corr}}$) is used according to Equation (3), which is presented below, to maintain a logical sequence of the equations used in the methodology. The detailed development of this correction is provided in Section 3.1.

$$m_{f_{Corr}}=2.219\times {10}^{25}\times {FFMC}^{-12.348}$$

(3)

Therefore, the FFMC is not calculated, and instead, ISI is determined using $m_{f_{Corr}}$ and windspeed according to Equation (4), resulting in “Initial Spread Index corrected” (${ISI}_{Corr}$). This equation has been adapted from the original CFWIS [7].

$${ISI}_{Corr}={(e}^{0.05039W})\times (91.9e^{-0.1386\times m_{f_{Corr}}}\times {\displaystyle \frac{1+m_{f_{Corr}}^{5.31}}{4.93\times {10}^7}}) ,$$

(4)

where $W$ is 10 m open windspeed in kilometers per hour and $m_{f_{Corr}}$ is the proposed fine fuel moisture content correction from Equation (3).

Finally, the “Fire Weather Index corrected” (${FWI}_{Corr}$) is determined using ${ISI}_{Corr}$ and BUI according to the Equation (5) (adapted from the original CFWIS [7]). The BUI is determined by DMC and DC, both based exclusively on original equations defined in [7].

$$\mathit{ln}\left({FWI}_{Corr}\right)=2.72\times {(0.434\times \mathit{ln}\left(0.1\times {ISI}_{Corr}\times BUI\right))}^{0.647}.$$

(5)

Figure 5 presents the flowchart of this methodology. In this figure, blue lines indicate the use of only meteorological data (original equations), while green lines indicate the incorporation of field measurements through moisture correction.

Days with FFMC lower than 82 (n = 1659) must also be determined due to the cumulative component of the CFWIS [7]. However, only days with FFMC ≥ 82 are considered (n = 1993), where for these days, the original indices (ISI and FWI) and corrected indices (${ISI}_{Corr}$,${FWI}_{Corr}$) are compared and related to fire activity.

2.3.3. Assessment of the Indices with the Fire Records

The methodology to assess the indices with the fire records is based on the previously presented percentile-based approach. In this case, we add to the number of fires (NF) and the burned area (BA) for the days considered (n = 1993). The probability classes were set in eleven, following the same distribution used by [38,56] to relate indices with fire records. The original indices (ISI and FWI) and the indices using moisture corrections of PP (${ISI}_{Corr}$, ${FWI}_{Corr}$) were related to the NF and BA.

3. Results

3.1. Correcting the Fuel Moisture Content Estimation

We applied a methodology based on a percentile approach to correct the fuel moisture content, as presented in the methodology. Figure 6 presents the relationship between $m_{f_{FFMC}}$ and FFMC and the proposed relationship between $m_{f_{PP}}$ and FFMC for the following probability class distributions: (a) n = 11, (b) n = 25, and (c) n = 40. The best fit between the variables was found using a nonlinear regression.

Table 1. Model summary for the nonlinear regression per number of probability classes (n): $FFMC=b·{m_{f_{Corr}}}^a$, where FFMC is the Fine Fuel Moisture Code, $m_{f_{Corr}}$ is the moisture content observed for Pinus pinaster, and “a” and “b” are the parameter estimates. “R²” is the correlation coefficient.

n	Parameters Estimates * Std. Error		R2	Std. Error of the Estimate
	a	b (Constant)
11	−12.955 * 0.482	$1.851\times {10}^{26}$ * $4.001\times {10}^{26}$	0.908	0.049
25	−12.348 * 0.563	$2.219\times {10}^{25}$ * 3.075 $\times {10}^{25}$	0.954	0.084
40	−12.138 * 0.502	$4.767\times {10}^{24}$ * 1.074 $\times {10}^{25}$	0.939	0.096

* Standard error associated with the parameter estimates.

presents the model summary for power law (correlation coefficient R², standard errors, and parameter estimates) for the proposed relationships.

The best fit for the data between $m_{f_{PP}}$ and FFMC was achieved with the power law, allowing for the precise identification of critical dryness conditions (low $m_f$ values). There are no significant differences in the relationship for each distribution of the number of classes (Appendix A, Figure A4). The model for the 25 classes was selected (Equation (3)) since it represents a sufficient number of points to create the fuel moisture correction and presents the highest correlation coefficient (R² = 0.954). This presents the fine dead fuel moisture content correction ($m_{f_{Corr}}$) based on field measurements of dead needles of Pinus pinaster made in Lousã (central region of Portugal). The minimum limit for applying this correction ($m_{f_{Corr}}\le 28.50\%$) corresponds to an FFMC of 82.

3.2. Indices Using the Correction of Fine Fuel Moisture Content

The fine dead fuel moisture correction was used to determine the corrected indices: ${ISI}_{Corr}$ and ${FWI}_{Corr}$ according to the methodology defined in Section 2.3.2. Using the Mann–Whitney U test, we found significant differences between original indices and corrected indices. The fine fuel moisture correction led to significantly higher values of ${ISI}_{Corr}$ and ${FWI}_{Corr}$ than those using the original equations (p-value < 0.001). The outputs of the statistical tests are presented in

Table A5. Output of the Mann–Whitney U test between corrected indices (${ISI}_{Corr}, {FWI}_{Corr}$) and original indices (ISI and FWI): (a) ranks, (b) test statistics.

(a)
Ranks
	${\mathit{m}}_{\mathit{f}}$	n	Mean Rank	Sum of Ranks
ISI	Corrected	1993	2127.08	4,239,274
	Original	1993	1859.92	3,706,817
	Total	3986
FWI	Corrected	1993	2073.99	4,133,471
	Original	1993	1913.01	3,812,620
	Total	3986
(b)
Test Statistics a
		ISI		FWI
Mann–Whitney U		1,719,796		1,825,599
Wilcoxon W		3,706,817		3,812,620
Z		−7.378		−4.416
p-value Asymp. Sig (2-tailed)		<0.001		<0.001

^a. Grouping variable: corrected and original.

(Appendix A).

The original indices and corrected indices were compared to observe the dispersion between them. The results are presented in Figure 7, showing (a) original ISI and corrected ISI and (b) original FWI and corrected FWI. Appendix A,

Table A6. Summary of linear regression (y = ax + b) between original indices and corrected indices: correlation coefficient (R²) and parameter estimates (a and b).

Original (x)	Corrected (y)	R2	Std. Error of the Estimate	Parameters Estimates * Std. Error
				a	b
$ISI$	${ISI}_{Corr}$	0.985	0.723	1.250 * 0.005	−0.367 * 0.033
$FWI$	${FWI}_{Corr}$	0.995	1.474	1.151 * 0.003	–0.344 * 0.058

* Standard error associated with the parameter estimates.

, provides a summary of linear regression between these variables.

The original indices and corrected indices exhibit a strong correlation. Both relationships show that the corrected indices (${ISI}_{Corr}$ and ${FWI}_{Corr}$) are systematically higher than the original indices (ISI and FWI) for the same values. For very low values of the original indices, the corrected indices may be slightly lower. The correction of the moisture content ($m_{f_{Corr}}$), instead of the moisture determined in the FFMC routine, resulted in indices that are more sensitive to variations in fuel moisture as the achieved values are higher. This suggests that the correction adjusts the indices to more accurately reflect the dryness conditions of the fuel.

3.3. Indices and Fire Activity

For this analysis, the data were arranged into 11 probability classes (Appendix A,

Table A7. Data arranged into 11 probability classes: number of days per class (n) for a total number of days ($d_{Total}$) of 1993, mean values of original indices (ISI and FWI), mean values of corrected indices (${ISI}_{Corr}$, ${FWI}_{Corr}$), mean values of number of fires (NF), and mean values of burned area (BA).

Probability Class	$\mathit{n}$	$\mathbf{Mean} \mathbf{of} {\mathit{m}}_{{\mathit{f}}_{\mathit{M}\mathit{e}\mathit{t}\mathit{e}\mathit{o}}}$ (%)	$\mathbf{Mean} \mathbf{of} {\mathit{m}}_{{\mathit{f}}_{\mathit{C}\mathit{o}\mathit{r}\mathit{r}}}$ (%)	$\mathbf{Mean} \mathbf{of} \mathit{I}\mathit{S}\mathit{I}$	$\mathbf{Mean} \mathbf{of} \mathit{F}\mathit{W}\mathit{I}$	$\mathbf{Mean} \mathbf{of} {\mathit{I}\mathit{S}\mathit{I}}_{\mathit{C}\mathit{o}\mathit{r}\mathit{r}}$	$\mathbf{Mean} \mathbf{of} {\mathit{F}\mathit{W}\mathit{I}}_{\mathit{C}\mathit{o}\mathit{r}\mathit{r}}$	Mean of NF	Mean of BA (ha)
[0.00–0.10]	199	18.35	18.61	2.47	4.76	2.39	4.55	0.20	0.06
]0.10–0.20]	199	16.83	17.16	2.92	5.67	2.80	5.44	0.21	0.04
]0.20–0.30]	199	15.23	15.55	3.51	8.61	3.37	8.33	0.27	0.09
]0.30–0.40]	200	13.92	13.39	4.28	12.88	4.60	13.62	0.46	0.17
]0.40–0.50]	199	12.81	11.60	4.79	15.44	5.64	17.46	0.54	0.10
]0.50–0.60]	199	11.89	10.29	5.43	18.67	6.74	21.82	0.64	0.27
]0.60–0.70]	200	10.91	9.06	6.06	21.03	7.79	25.11	0.89	1.93
]0.70–0.80]	199	9.78	7.83	7.25	26.28	9.48	31.51	0.96	2.61
]0.80–0.90]	199	8.40	6.54	9.16	30.58	11.82	36.23	1.19	9.17
]0.90–0.95]	100	6.91	5.38	10.69	36.04	13.20	41.34	1.21	82.33
]0.95–100]	100	5.18	4.32	13.81	46.03	15.42	49.44	1.96	1129.18
	$d_{Total}=1993$

). The original indices and corrected indices were related to the number of fires (Figure 8a,c) and burned area (Figure 8b,d) using average values of these variables based on the percentile methodology. In the figures showing BA, the vertical scale (y) is logarithmic.

Table 2. Summary of nonlinear regressions between indices and fire activity (number of fires—NF; burned area—BA) in average terms. For NF, the model used was the Power law (y = bx^a), and for BA, the model used was Exponential (y = be^a.x). “R²” is the correlation coefficient, and “a” and “b” are the parameter estimates.

Index (x)	Fire Activity (y)	R2	Std. Error of the Estimate	Parameters Estimates * Std. Error
				a	b
$ISI$	NF	0.964	0.152	1.368 * 0.088	0.058 * 0.009
${ISI}_{Corr}$	NF	0.984	0.102	1.179 * 0.050	0.069 * 0.007
$FWI$	NF	0.982	0.106	1.011 * 0.045	0.036 * 0.005
${FWI}_{Corr}$	NF	0.977	0.121	0.917 * 0.047	0.043 * 0.006
$ISI$	BA	0.975	0.545	0.920 * 0.049	0.003 * 0.001
${ISI}_{Corr}$	BA	0.948	0.791	0.727 * 0.057	0.005 * 0.002
$FWI$	BA	0.957	0.721	0.246 * 0.017	0.007 * 0.003
${FWI}_{Corr}$	BA	0.930	0.920	0.211 * 0.019	0.009 * 0.005

* Standard error associated with the parameter estimates.

presents the parameter summary of the nonlinear regressions between the indices (original and corrected) and fire activity. For both ISI and FWI, the best fit with NF and BA was found using the power law and exponential law, respectively.

Overall, indices show a strong positive correlation (R²) with NF and BA in average terms. The indices relate better to the NF than to the BA. Usually, the indices exhibit broadly equivalent discriminatory power for fire occurrence, as also found by [57], although a lower correlation with BA is typically observed in other studies due to various factors (wind speed, number of simultaneous fires, firefighting efficiency, fuel nature, topography) [29]. The exceptional nature of the events of 2017 in relation to the normal or average situation of the region is evident in the BA, which highly increased the last percentile (P > 95).

The higher R² for the relationship between ${ISI}_{Corr}$ and NF (Figure 8a) suggests that the developed correction may provide a slightly more accurate prediction for the NF than the original ISI. However, for most other relationships with NF and BA (Figure 8b–d), the original indices had slightly higher correlation coefficients than the corrected indices. Thus, the original indices can already capture the variation well enough to predict NF and BA effectively.

4. Discussion

Pinus pinaster (PP) and Eucalyptus globulus litter are common components of fuel beds in Central Portugal and the Mediterranean basin, and they are known for their significant role in wildfire ignition and spread. Despite the abundance of both species and the availability of data for both of them, this study focused exclusively on PP litter following [38] due to the absence of significant practical differences in moisture content compared to Eucalyptus globulus. Given the importance of drier conditions in the assessment of fire danger, we selected the days with a moisture content of dead needles of PP below 28.50%, aligning with fire managers who consider this value as the threshold for potential fire spread [35].

Fine dead fuel moisture estimated by the FFMC ($m_{f_{FFMC}}$) is significantly higher than the moisture content of field measurements of PP ($m_{f_{PP}}$) (p-value < 0.001) as also observed by [40]. This difference is partly influenced by the topography and its south solar exposure where field measurements were conducted [25], but this condition does not impact the fine dead fuel moisture content estimated by the CFWIS, which assumes flat terrain. According to [25], weather stations provide limited insight into the variability in fuel moisture in complex topography due to their failure to sample a range of terrain conditions.

The development of a correction equation to more accurately estimate the fine fuel moisture content ($m_{f_{Corr}}$) based on field measurements of dead needles of Pinus pinaster is a contribution to the new relationships that are often established between one or more of the six components of the CFWIS [31]. Applying the $m_{f_{Corr}}$ (Equation (3)) according to the methodology proposed (Figure 5), resulted in corrected indices (${ISI}_{Corr}$ and ${FWI}_{Corr}$) that present higher and potentially more accurate values under extreme dry conditions (Figure 7). Understanding the variations of fine fuel moisture content using the correction equation that more accurately reflects local conditions allows fire managers to refine their strategies for resource allocation. Specifically, days with $m_{f_{Corr}}$ lower than 4.32% (Table A7) correspond to extremely dry conditions. On such days, fire management teams could allocate more resources to areas with these conditions and enhance early detection efforts. It is important to note that these results are particularly significant for days with extreme danger conditions, where the fine fuel moisture content may be underestimated by the original CFWIS. Moreover, improved methods for predicting fire danger provide timely and accurate information, which is crucial for effective communication and preparedness and helps to minimize the impacts of wildfires on communities and ecosystem services.

The analysis of fire activity (Figure 8) provides valuable indications about the relationship between indices and the number of fires (NF) and burned area (BA). Although the corrected indices showed higher sensitivity under dry conditions, they did not demonstrate a consistent improvement in predictions of the NF and BA compared to the original indices (ISI and FWI). These results indicate that the CFWIS is sufficiently robust to variations in the moisture content of fine fuels in a certain region. The correction proposed for the central region of Portugal improves the sensitivity of the indices under dry conditions but does not provide a significant improvement in the prediction of fire occurrence and burned area compared to the original indices.

Although it is beyond the scope of this study, it is worth mentioning that while there is no significant difference between calculating the indices using the original or corrected methods in relation to fire activity, interpreting the FWI values must consider regional historical fire data to define daily fire danger levels accurately. For similar weather conditions, the same FWI value can indicate different danger levels in different regions [38]. Thus, while the CFWIS is robust, its FWI outcome must be regionally calibrated to provide accurate fire danger classes, as demonstrated in studies such as those by [38,55,56,58].

A limitation inherent to this study was the use of a single weather station to determine the indices and subsequently relate them to fire records for the entire region (2010 km²). While this approach has been followed by other authors [48,49], it can be argued that this method, though attractive in its simplicity, may not accurately represent the average meteorological conditions across the entire region. Therefore, we recommend the use of data from multiple weather stations whenever possible.

The following points summarize the implications of the present study and provide suggestions for its continuation:

• This methodology can find applicability in countries with their own dataset of field measurements. Alternatively, the correction proposed for the fine fuel moisture content (Equation (3)) can be applied using the local meteorological parameters and the fire history of the region;
• Future research could explore establishing relationships for field measurements of moisture content exceeding 28.50%;
• Additional studies should be carried out for layers composed of different fuels represented by DMC and DC to obtain proper corrections and verify the performance of those corrections with fire activity;
• Simultaneously, field measurements made in Lousã can serve as an indicator of fire danger in a large area of Portuguese territory [23], providing valuable knowledge for comparison with other ecosystems.

5. Conclusions

This study contributes to increasing knowledge about using field measurements to empirically calibrate the indices of the Canadian Fire Weather Index System (CFWIS), calculated for a specific region. A dataset of field measurements of fine dead fuel moisture content (dead needles of Pinus pinaster) lower than 28.50% conducted in Lousã, in the central region of Portugal, spanning a decade from 2014 to 2023, was used. A correction to determine the fine dead fuel moisture content based on fine fuel moisture content (FFMC) and field measurements was proposed.

ISI and FWI using the moisture correction (corrected indices) are significantly higher in dry conditions than ISI and FWI based exclusively on meteorological data (original indices). However, when we relate them to fire activity, we see that the corrected indices and original indices have similar behavior. While it is often suggested that adapting an existing fire danger rating system for a new environment requires developing new relationships or modifying existing ones, it is worth considering that such adaptations are not always required. Our results suggest that CFWIS is sufficiently robust to handle variations in moisture content without the need to modify the indices equations.