FIRE HAZARD FORECAST BY THE REGIONAL CLIMATE CHANGE PROJECTION USING THE ETA MODEL: A CASE STUDY IN BAHIA, BRAZIL* PREVISÃO DE RISCO DE INCÊNDIO PELA PROJEÇÃO DE MUDANÇAS CLIMÁTICAS REGIONAIS USANDO O MODELO ETA: UM ESTUDO DE CASO NA BAHIA, BRASIL

This article proposes a method for predicting fire occurrence, considering regional climate change projection using the Eta model, with a 20 km resolution, for the RCP4.5 and RCP8.5 scenarios. Fire occurrence in the state of Bahia was calculated as a function of the three main sensitivity factors on a daily time-scale: days without precipitation, precipitation, and maximum temperature. Historical fire occurrences from 1998 to 2018 and meteorological data from 1960 to 2018 were obtained from official institutes, and weather forecast parameters from 2018 to 2050 were downscaled from the web platform PROJETA. The correlations between the meteorological factors and fire occurrence were calculated for the historical data and a weight factor corresponding to a control simulation. Afterwards, a correction factor was determined, based on the historical fire occurrence data used for the forecast in the two scenarios. The results indicate that between 2018 and 2050, risk of fire will have an average increase of 27% at the RCP4.5 and 38% at the RCP8.5 scenario.


Introduction
Wildland fire is, in many cases, an essential ecosystem component that ensures the sustainability of its processes and communities. Wildland fire plays a key role as an environmental filter, selecting for species and their traits, and shaping ecosystems' communities (Aponte et al., 2016). However, when there are changes in climate alone, they may have the potential to alter the distribution of vegetation types within the region, and climate-driven shifts in vegetation distribution are likely to be accelerated when coupled with standreplacing fire (Hurteau et al., 2014). Also, it has been shown that the frequency and intensity of wildland fires increase over the coming decades. Then, efforts to fully understand the implications of this growth will assist decision-and policy-makers to develop a more comprehensive understanding of the impact of them, and therefore the benefits of reducing the incidence of fires through GHG mitigation (Lee et al., 2015).

According to the Intergovernmental Panel on Climate
Change (IPCC), climate change will likely increase the global risk of extreme fire events (Stocker et al., 2013).
Throughout the century, situations of hot winds and even hotter anticyclonic events will arise and increase, resulting in an increase of the potential of large fire events (Duane et al., 2019). In general, climate conditions are a fundamental driver of fire spread, and fire patterns are strongly sensitive to regional climate variability and (Silva et al., 2016;Eugenio et al., 2019).
Fires primarily occur after prolonged dry spells where the air temperature is high and climate change will produce conditions more conducive to severe fires (Stephens et al., 2020). Dry vegetation can then be easily ignited, starting a wildfire that quickly spreads out of control with strong winds. Most fires near populated areas are caused by human activity, while a smaller portion occur naturally because of lightning events. Alongside accidental fires, a significant number are also started deliberately. For the risk hotspot regions, models can be used to estimate occurrence of these events. However, because of several man-made and natural factors, modelling this risk is complex and influenced by subjective assessments. The risk is especially high in climate zones where there is enough rainfall to allow vegetation to flourish part of the time, yet also have long periods of warm weather with little precipitation. Under these conditions, plants gradually dry out and become highly flammable. This has been reported in different regions of the world like Asia, North America, Europe, and the United States (Liang et al., 2017;Duane et al., 2019;Halofsky et al., 2020;Stephens et al., 2020;Vilà-Vilardell et al., 2020). Brazil is exposed to a variety of natural hazards, such as droughts and excessive rainfalls, which are the most frequent and damaging events. Fire is one of the most important types of disturbance affecting forest landscape ecosystems, especially in semiarid biomes and grassland.
There are some locations in Brazil, such as Mato Grosso State, which a large part of its area (55.06%) is under high to extreme risk of fires (Mota el al. 2019) (Ziccardi et al., 2018). Fire corresponds to the classification of disasters related to the intense reduction of water precipitation.
This phenomenon can occur due to natural causes as well as human actions, such as climatic and environmental factors, which are decisive for increasing the rate of fires. The frequency and distribution of fires in Brazil are strongly associated with climatic conditions; the increase in temperature and changes in seasonal and annual rainfall have a large influence on fire occurrence  study of finer-scale processes, as is the case of fire. In this regard, a strong effort has been made by the scientific community to provide refined information about future climate using Regional Climate Models (RCMs).
The present study aims to assess the future fire hazard in the Brazilian state of Bahia, using a correlation index of meteorological aspects and fire occurrence in a regional regime. For this purpose, a regional downscaling of meteorological aspects was made by the PROJETA platform of the regional model. An assessment of the two RCM models has been calculated to indicate the fire risk change over a Brazilian region between 2018 and 2050.

Study Area
Most of the fires in Brazil that occur in the North and     According with Torres et al. (2017), one of the factors that explain the fire occurrence is the land decline, being as higher the slop e of the terrain. The Chapada Diamantina National Park (PNCD), an integral protection unit, suffers recurrent fires and represent the highest number of fires recorded among the federal integral conservation units (dos Santos et al., 2020;Franca-Rocha et al., 2017).
Every year the region of Bahia suffers from a large number of forest fires, which devastates the local fauna and flora. Forest fires are common in Bahia during the dry season, which runs from July to October.

Methodological approach
The method is focused on fire hazard occurrence by quantitative and qualitative data analysis from four sources: temperature. The fire occurrence forecast was obtained from the equation function of meteorological projections data, past fire occurrence, and past meteorological data. The method is divided into five calculation steps as described below: • Firstly, the annual correlations between the historical fires' occurrence and meteorological historical data between 1998 and 2018 (Equation 1) was calculated.
Correlation i (x, y i ) =∑ (x -x') (y I -y i ') / (√(x -x') 2 (y i -y i ') 2 ) x = fire occurrence y i = meteorological data The correlations were calculated between the two indexes to access annual variability. The statistical evaluations consider the relationship between a dependent (fire occurrence) and an independent variable (meteorological data) to calculate the determination coefficient.
• Secondly, based on historical data, for each correlation factor and the equation for fire occurrence (Equation 3) using a matrix method for three variables, the correspondent weight factor (α, β, µ) was calculated (Equation 2).
Weight factor i = C 1 / ∑ (C 1 + C 2 + C 3 ) • Thirdly, the historical fire occurrence was calculated while considering three main sensitive historical meteorological factors: days without precipitation (positive relation), precipitation (negative relation), and maximum temperature (positive relation) between 1998 and 2018 (Equation 3).
Fire occurrence Historical = ∑ f (α * C 1 -β * C 2 + µ C 3 ) C 1 = correlation with days without precipitation α = correspondent weight factor C 2 = correlation with precipitation β = correspondent weight factor C 3 = correlation with maximum temperature µ = correspondent weight factor • Fourthly, in order to calculate a correction factor for each scenario, a control simulation was developed, and the results were equated with the real historical fire occurrence data and fire forecast from 2006 to 2018 (Equation 4).
Correction factor i = median (historical number of fire occurrence / projection number of fire occurrence) • Finally, the average of the correction factors was used for the forecast in the two scenarios RCP4.5 and RCP8.5 between 2018 and 2050 (Equation 5).
Fire occurrence Forecast = ∑ (Fire occurrence Historical ) * correction factor The climate impact on fire risk results from the interaction between climate-related hazards (including hazardous events and trends) and the vulnerability and exposure of human and natural systems. A risk associated with climate depends of two main factors: vulnerability and exposure (IPCC, 2012). Therefore, this is the very first analysis, which is based on comparing the RCP 4.5 and RCP 8.5 scenarios and the assumption that the other aspects will not change until 2050, in order to forecast fire hazards.
The Representative Concentration Pathways (RCPs) of IPCC are scenarios that include time series of emissions and concentrations of the full suite of greenhouse gases (GHGs), aerosols, chemically active gases, and land use/ land cover (Moss et al., 2008). The word representative signifies that each RCP provides only one of many possible scenarios that would lead to the specific radiative forcing characteristics.
Simulations were performed in the Bahia region over South America. The downscaling was done of monthly values of maximum surface temperature, days without precipitation, and precipitation that were all extracted from the selected study area considering two scenarios simulations: RCP4.5 and RCP8.5 performed for the years 2018-2050, these are: • RCP4.5 an intermediate stabilization pathway in which radiative forcing is stabilized at approximately 4.5 Wm -2 and 6.0 Wm -2 after 2100 (the corresponding ECPs assuming constant concentrations after 2150); • RCP8.5 one high pathway for which radiative forcing reaches levels greater than 8.5 Wm -2 by 2100 and continues to rise for some amount of time (the corresponding ECP assuming constant emissions after 2100 and constant concentrations after 2250).

Web platform PROJETA
The web platform PROJETA, which is an acronym for "Projections of climate change for South America downscaled by the Eta model", was built to automatically access, prepare, and make the dataset of the downscaling climate change scenarios available to users. These projections are based on global climate downscaling models carried out by the Eta Model at CPTEC/INPE. The PROJETA project is a partnership between CPTEC/INPE and the University of Passo Fundo, promoted by the Brazilian Ministry of the Environment and funded by the German agency Deutsche Gesellschaft für Internationale Zusammenarbeit.
The premise of PROJETA is the automatization of the extraction process and the availability of data from regionalized climate projections for Brazil. This allows broad and unrestricted access to various available climate parameters and aims to meet access demands for climate projection data, treated and compatible with sectoral analysis programs and platforms. To obtain data on the web platform, the authors first choose the model RCP and climate scenario 4.5 and 8.5 with a resolution 20km. Afterwards, frequency selection is done to the database daily, through 2018-2050, using more than 30 years for this analysis. Later, the variables (days without precipitation; precipitation; maximum temperature) were selected.
Finally, the area of interest of analysis is selected, in this case the state of Bahia. The generated data was then sent by PROJETA to the authors. The data received was statistically treated to allow the analysis.

Results and discussion
The quantitative results of historical fire data analysis, taken  (Santos, 2004). The results of the qualitative data indicated the fire occurrence is mainly related to climate conditions (drought and with high temperatures), associated with some human activities caused accidentally (e.g. road accident) or intentional (e.g. manage pastures).

Forecast of fire occurrence
The results of the fire occurrence forecast indicate an average increase of 27% at the RCP4.5 and an increase of 38% RCP8.5 scenario for the state of Bahia (fig. 5). The historical data shows a decrease of events, which may be related to climate and public policy factors such as the creation of the program Bahia Without Fire in 2010 (SEMA, 2020).

Evolution of fire hazard
The results indicate that, fire occurrence is observed and increasing, overall the fire hazard had a significant increase of 21% comparing the RCP 4.5 and RCP 8.5 scenarios in the state of Bahia (this value was calculated considering the total sum of the annual difference of fire occurrence). In 2033, the maximum increase of 166% is observed ( fig. 6).

Conclusions
The analysis of the scenarios was able to reproduce the main meteorological patterns and historical data on a regional scale that allowed the evaluation of the sensitivity of fire hazard changes expected until 2050, which would increase over time and could be used in the future for assessing a complete risk assessment. This study shows a systematic increase in the fire hazard that is observed in the Brazilian region of Bahia. Therefore, the method chosen seems to be appropriate since it allowed the authors to obtain results using available tools for downscaling data on a    -50% 0% 50% 100% regional scale. However, it is important to mention that the method does not consider variables such as public policies and other political decisions, which may influence the occurrence of fires in Bahia. However, the three climate variables and the web tool PROJETA enables the support of further risk assessments to propose and implement public policies for reducing fire hazards. To improve the fire forecast modelling, it is necessary to have more fire data available, since there is only data from 1998. According to INPE, the data from 1992 to 1997 is not provided as they are not compatible with current data (from 1998) that uses a different statistics treatment. However, as the result graphs indicated upward and downward peaks of fire outbreaks, a circular variation of the intensification and reduction period of the fires can be shown by 2050. Thus, it can help the planning and monitoring of related mitigative and preventative actions.