The influence of climate parameters on fires in the Paraíba do Sul River valley, southeast Brazil

Abstract Brazilian biomes have been experiencing an increase in fires during the whole year, but fires increase substantially during drier periods. Several indexes might be good indicators of the severity of the droughts, such as The Rainfall Anomaly Index (RAI), Standardized Precipitation Index (SPI) and the Vegetation Health Index (VHI). This study therefore aimed to understand the dynamics of climate, using some indexes and fires in Paraíba do Sul River Valley, Paulista portion, to verify whether fire is more likely to spread in hotter and drier years. We hypothesized that fire events are more frequent and burned areas are larger in hotter and drier years in the region. By conducting a cross-correlation analysis and separating the monthly data into dry and rainy seasons, it was possible to establish a correlation between climate parameters and fire. A significant correlation was found between RAI and fires in both seasons. Additionally, we observed that high occurrences of fire events and burning areas were more explained by RAI, VHI and SPI-3 in dry and wet seasons than by temperature and SPI-1, SPI-6 and SPI-12. We noticed a complex dynamic between fire events, burned area, the environment, and climatic variables. However, the studied indexes proved to be effective tools for detecting drought conditions in the region and their relationship to fire.


INTRODUCTION
Every year, wildfires burn more than 400 million hectares worldwide (Andela et al., 2017) and shape the structure and diversity of all biomes (Bond and Keeley, 2005).Since 2018, all Brazilian biomes have been experiencing an increase in fires (INPE, 2022).Humans are a major force driving many fire regimes around the globe (Archibald et al., 2013), and, in Brazil, fire is often a key process when considering the drivers of forest loss and commodities agriculture (Barlow et al., 2020).
Fire occurrence and propagation may be directly linked to climatic factors, such as precipitation, temperature and wind speed (Keeley et al., 2011).In Brazil, fires have been registered during the whole year, but burnings increase substantially during drier periods (INPE, 2022).Lately, the increase in fires has been related to longer duration of dry seasons worldwide (Jolly et al., 2015), explained by climate change (IPCC, 2021).According to Pivello et al. (2021), drier climates and land-use changes increase the risk of wildfires throughout Brazil.
The Paraíba do Sul River Valley is currently characterized by a landscape patchwork of tropical forest remnants and pastures (Sapucci et al., 2021).Guedes et al. (2020) found a strong positive correlation between pasture area and burn probability in the region, while the opposite was verified for forest cover, reinforcing the role of pastures in fire ignition and transmission.Several conservation and restoration initiatives taking place in the region might be damaged by fires (WRI, 2022: FLR Hub, TNC: Programa Conservador da Mantiqueira), but also economic activities and population health.The region presents a seasonal precipitation variability, with a well defined rainy and dry season, associated with the South America Monsoon System (Reboita et al., 2010).In winter (end of June to end of September), the accumulated precipitation does not exceed 200 mm and in summer the values reach 800 mm (Brasiliense et al., 2020).Ayres (2010) showed that floods and severe storms have a high frequency in the region.Zilli et al. (2016) analyzed extreme rainfall events in the rainy season across the southeastern coast of Brazil and, for the Paraiba do Sul River Valley Basin, they found an increase in the frequency and intensity of these events.Drought events are also recurrent in the region (Santana et al., 2020) and have a major impact on water supply as registered on 2013/2015 (Coelho et al., 2016;Marengo et al., 2015;Nobre et al., 2016).
Drought can be defined as a failure in hydrological balance.Such failure can include reduced rainfall over a period (more than average), inadequate timing of the precipitation, or a negative water balance due to increased atmospheric water demand related to high temperatures (UNDRR, 2021).Several indexes have been developed and are proven to be good indicators to measure the severity of the droughts around the world, such as the Rainfall Anomaly Index (RAI: Rooy, 1965), the Standardized Precipitation Index (SPI: Hayes et al., 2011), and the Vegetation Health Index (VHI: Tran et al., 2017).RAI, SPI and VHI have been adopted to identify drought events in southeast Brazil (RAI: Noronha et al., 2016;SPI: Blain and Kayano, 2011;Silva andMello, 2021 andVHI: Gomes et al., 2017).By using these indexes, we intend to have a broader analysis of drought, since SPI is based on a probability density function, with adjustable scales, RAI determines deviations referring to the normal condition of precipitation, with greater visible seasonality, and VHI is used as an indicator of the response of stressed vegetation.
In addition, in the last few years, studies have been applying drought indexes to verify the relation to wildfires occurrences in biomes and regions in Brazil: Alvarado et al. (2017), found increasing fire occurrence during the driest periods in Cerrado; Marengo et al. (2021) identified droughts by indexes such as VHI, concluding that water and heat stress increases flammability thresholds in Pantanal; and Teodoro et al. (2022) verified that the trend of fire occurrence in the Midwest, Southeast, and South regions of Brazil is related to precipitation, land surface temperature and SPI.
Fire risk assessment studies are extremely important to plan preventive measures, optimizing use of resources, compared to suppressive measures.This study therefore aimed to understand the dynamics between climate and fires in Paraíba do Sul River Valley, Paulista portion, and to verify whether fire is more likely to spread in hotter and drier years.We hypothesized that fire events are more frequent and burned areas are larger in hotter and drier years in the region.

Study site
The Paraíba do Sul River Valley, Paulista portion (46.5°W -43.5° W/ 22.5° S -24° S, Figure 1), is located in southeast Brazil, São Paulo State.The valley population of over 2 million people (IBGE, 2021) in 39 municipalities sits on the fringe between the major metropolitan areas of São Paulo and Rio de Janeiro.Although populous, the mostly rural valley is currently characterized by a landscape patchwork of tropical forest remnants and pastures (Sapucci et al., 2021).The region has a hilly relief, located between two mountain chains, "Serra da Mantiqueira" and "Serra do Mar" which altitude ranges from 800 m to 2,500 m above sea level, with its interior valley varying between 560 and 650 m altitude (Devide et al., 2014).

Datasets and index calculation
Monthly precipitation data was obtained for the studied region at Climate Hazards Group InfraRed Precipitation (CHIRPS, Funk et al., 2015), with a spatial resolution of 0.05° × 0.05° from January 1981 to December 2021.For air temperature, monthly grid point data were used from the Climate Research Unit dataset (CRU, Harris et al., 2020), with a spatial resolution of 0.5º × 0.5º, from 1981 to 2020 (Table 1).Fire data included monthly fire occurrences available at Programa Queimadas (INPE, 2022), from 1998 to 2021, with a spatial resolution of the sensors image AVHRR (NOAA satellites), MODIS (AQUA and TERRA satellites) and ABI (GOES satellite) about 1km x 1km, and burned area, with 300-meter resolution, available at Copernicus Global Land Service platform (CGLS, Hagolle et al., 2005), through the 10 day composite burns detected.Although the used fire dataset is not homogeneous due to the different sensors and satellites, it is the most extended available dataset (from 1998), so it was selected for this study.Built-in NOAA, TERRA/AQUA, and GOES detection allow the monitoring of all focuses accumulated throughout the day by these three types of satellites.Thus, for precipitation and temperature data, a spatial average for the study area was considered for each year of the studied period.The fire data were also accounted for, as point data, in the area.says that a drought episode can be noticed when there are at least two consecutive months with SPI values lower than -1.
RAI verifies deviations referring to the normal condition of precipitation (Rooy, 1965), allowing comparison to a time series or spatiality evaluation of a drought.Therefore, measures are allocated in descending order and the average of the ten highest values is calculated, followed by the average of the ten lowest, which are represented by M ̅ and X ̅ , respectively, as shown in Equations 2 and 3, since N is the monthly or annual precipitation (mm) and N ̅ is the average monthly or annual precipitation of the series (mm).
Equation 2 expresses index value for positive anomalies, while Equation 3 refers to negative anomaliesbelow average.RAI was calculated with point data, on monthly and annual scales, and its reference values are > 4, for extremely rainy; 2 to 4, for very rainy; 0 to 2, for rainy; 0 to -2, for dry; -2 to -4, for very dry; < -4, for extremely dry (Araújo et al., 2009).
VHI characterizes the health of the vegetation by assuming that stressed conditions are linked to lower than normal NDVI and higher than normal temperature (Kogan, 2001); it may be written as Equation 4.Where VCI refers to Vegetation Condition Index (Equation 5); TCI to Temperature Condition Index (Equation 6); T to the smoothed weekly temperature; Tmax and Tmin to its maximum and minimum, respectively; NDVI to the smoothed weekly index, NDVImax and NDVImin to its maximum and minimum.TCI is given by thermal channels of AVHRR (Kogan, 1995); thus, its temperatures depend mainly upon the surface temperature, but also on the total-column atmospheric water vapor, and on the surface-atmosphere temperature gradient (Gutman et al., 1995).VHI reference values are 30 to 40, for abnormally dry; 20 to 30, for moderate drought; 12 to 20, for severe drought; 6 to 12, for extreme drought; < 6, for exceptional drought (Cunha et al., 2019).According to CEMADEN (2021), if the VHI returns a value lower than 40 and this scenario persists for two consecutive months, a drought event is verified.This condition remains in effect until the index value reaches at least 45.We separate the climatic and fire data according to seasonality (wet, from October to March, and dry seasons, from April to September), and after that, to analyze whether SPI, RAI, VHI, 90th percentile for temperature and 10th percentile for rainfall (explanatory variables) were related to burning events and size area (response variables), we performed Pearson correlation tests.By using different indexes, it was possible to understand not only the drought event on a given time scale, but also the drying of vegetation and the deviations from the normal precipitation condition.The tests were considered statistically significant when p-value was lower than 0.05.Analyzes were performed in R Version 3.6.3(R Core Team, 2019), by using the SPEI package for SPI calculation.

RESULTS AND DISCUSSION
The monthly variability of precipitation (1981 -2021) and temperature (1981 -2020) can be described in terms of seasonality, i.e., climatology is represented in blue in Figure 2. The higher accumulated rainfall occurred in summer months (234.23 mm were the average recorded in December; 269.60 mm in January; and 223.66 mm in February) and the lowest values in Winter (41.34 mm of rain in June; 33.30 mm in July; and 34.55 mm in August).Autumn (end of March to end of June) and Spring (end of September to end of December) were transition seasons with average precipitation between 64.47 mm (May) and 165.79 mm (November).The highest mean temperature occurred in February (22.8ºC) and the lowest in July (16.02ºC).The average annual precipitation was 1574.7 mm (standard deviation: 174.0 mm) and the average mean temperature was 19.9°C (standard deviation: 0.5°C).

SPI, RAI and VHI temporal variability
We found a temporal variability in SPI indexes, ranging from dry to wet periods (Supplementary Material 1) in Paraíba do Sul River valley.We found 12 drought episodes (Figure 3) from 1998 to 2021 in Paraíba do Sul River Valley.October 2019 to December 2020 (14 months) was the longest drought event, when severe drought was registered for three months scale, and extreme, for six months scale.Further, 2003 was a dry year regarding duration and severity, and were also, 1999, 2000 and 2018; while the first two were extreme The influence of climate parameters on fires … Rev. Ambient.Água vol.18, e2923 -Taubaté 2023 and the last, severe drought.Considering the entire series, the lowest SPI-3 values were in March 1984 (-2.82), June 2000 (-2.47) and November 1999 (-2.41).By calculating the ratio between drought severity (sum of SPI values, on its scale of 1, 3, 6 or 12 months, in the drought period, Fernandes et al., 2009) and duration in months, we found the most intense drought years to be 2018, 2000 and 1999.RAI was also seasonal (Supplementary Material 2), with the driest months concentrated on June, July and August when very dry or dry categories were registered for the entire series.Noronha et al. (2016) observed the same results to the northwest of Rio de Janeiro state.The same authors stated that the seasonal behavior of the index and the great temporal variability of precipitation makes it difficult to calculate a possible anomaly in a small time interval.In general, between April and September, at least 85% of the years in the series had RAI ≤ -0,5 (dry).Only 1984 was classified as an extremely dry year and, after 1998, five years were classified as very dry, 2007, 2003, 1994, 1999 and 2014.Regarding VHI, the longest drought since 1998 was from December 1999 to September 2000 (10 months), followed by June to October 2003, which overlapped SPI results and from August to November 2017, as shown in Figure 3 (and Supplementary Material 2).Thus, even though vegetation health is associated with precipitation conditions, not all SPI scales were linked to that.That might be explained by different vegetation types and how they respond to drought; i.e., denser vegetation may be more prone to burning when higher scales of SPI point to a long-term accumulated rainfall deficit, while pastures may have higher susceptibility due to a shorter period of water deficit.

Fire temporal variability
Highest proportions of burned areas (46.4%) occurred in July and August, and 37.4% occurred in Autumn.The same seasonality was observed for the number of fire occurrences, with 43.8% of burnings being recorded in Winter, followed by 27.8% in mid-spring.In addition, we noticed a maximum number of fires (157) in August 2003, and of burned areas (409.71 km²) in July 2021, as shown in Figure 4. August and September were also months with the highest susceptibility to fires in the remnants of the Atlantic Forest biome in Rio de Janeiro state (Clemente et al., 2017).We found above average fire events in Summer and Winter and above average fire areas in Winter (Table 2).Winter (end of June to end of September) 30% 100% Spring (end of September to end of December) 25% 40%

Relation drought x fire
We found that precipitation, RAI, VHI and SPI-3 had more significant relations with burning events and area in dry and wet seasons (Table 3) than temperature and SPI-1, SPI-6 and SPI-12.
Lower precipitation explained more fire events and greater fire size in dry and wet seasons (Table 3), while temperature was not related to fire in the region (except for fire events in the wet season, as shown in Table 3).In Paraíba do Sul River valley, pasture is the main land use (Sapucci et al., 2021) and cattle ranchers use fire (illegally) as a management tool to increase productivity of pastureland.Burnings that happen in dry months usually become large wildfires, due to climatic conditions, affecting large extensions and native forests in the region (Guedes et al., 2020).
In general, SPI (except for SPI-3) was slightly related to fire in Paraíba do Sul River valley (Table 3).Similarly, other studies have found weak correlations between SPI and fire, such as for abnormal weather conditions influenced by El Niño and La Niña events.SPI-3 was associated with fire events (dry and wet seasons) and fire size in the dry season.RAI related to fire events and fire size in dry and wet seasons (Table 3).2003, classified as a very dry year, according to RAI, had the highest number of fires in the period (489), while 2014 exceeded the average of burned area for subsequent yearsdisregarding years with possible absence of data -(143.2km²), burned from April to December 2014.VHI is associated with fire events in both wet and dry seasons and fire size in the wet period (Table 3).
Fires in dry and wet seasons were explained by almost the same range of climatic variables, the only differences were: temperature and fire events in the wet season; SPI-1 and fire events in the dry season; VHI and fire size in the wet season and SPI-3 and fire size in the dry season (Table 3).

CONCLUSIONS
This study aimed to understand the dynamics between climate properties and fires in Paraíba do Sul River Valley, Paulista portion, and to verify whether fire was more likely to spread in hotter and drier years.Although we did find fire events being more frequent and burned areas being larger in hotter and drier periods in the region, we also noticed a complex dynamic between fire events, burned area and climatic variables.Nunes et al. (2015) had similar conclusions and pointed that higher linear correlation among the adopted parameters is found in more detailed analyses of local sites.
Specifically, we found that high occurrences of fire events and burning areas were more explained by RAI, VHI and SPI-3 in dry and wet seasons than by temperature and SPI-1, SPI-6 and SPI-12.In this context, the indexes proved to be good instruments for detecting drought conditions in the region.This study can contribute to preventive measures regarding fire susceptibility in the landscape.

Figure 1 .
Figure 1.Study area indicating the Paraíba do Sul River Valley, Paulista portion, in Brazil, South America.

Figure 3 .
Figure 3. Identification and classification of drought episodes in Paraíba do Sul River Valley, from 1998 to 2021, according to the SPI-3 (upper) and to VHI (lower).In the top figure, light purple is moderate, dark purple is severe and red is extreme drought.In the bottom figure, light purple is abnormally dry and dark purple is in moderate drought.The years 2012 and 2017 are repeated due to the occurrence of two drought events in each of them.

Figure 4 .
Figure 4. Annual distribution of occurrences of fires and burned areas over the years in Paraíba do Sul River Valley, since the burned area comprises the period from April 2014 to July 2021 and the number of fires from 1998 to 2021.Table 2. Percentage of coincidence of months considered climatic extremes of temperature (hot extreme) or precipitation (dry extreme) for each season and above-average fire events.Extreme conditions of temperature or precipitation at Nunes et al. (2015);Alvarado et al. (2017), for annual rainfall data, andTeodoro et al. (2022),The influence of climate parameters on fires … Rev. Ambient.Água vol.18, e2923 -Taubaté 2023

Table 1 .
Climatic and fire variables, and the respective data coverage periods, databases and references for each database.

Table 3 .
Correlation coefficient among climatic variables and fire events and burned area on dry and wet seasons.Bold represents P < 0.05 and significant correlations.