Nota Técnica n. 11

Health Impacts of Deforestation-Related Fires in the BrazilianAmazon∗

Andre Albuquerque Sant’Anna†and Rudi Rocha

IntroductionThis report aims to assess correlations between thedeforestation-related fires in the Brazilian Amazon andhealth outcomes, especially respiratory illness. In thisdocument, we discuss the methodology that will be uti-lized for the statistical study linking active fire hotspots,ambient air pollution and health. Finally, we provide themain results as well as aggregate estimates of the e�ectsof ambient air pollution related to fires and deforestationon health in the Brazilian Amazon. For the purposes ofthis report, we define the Brazilian Amazon as the Ama-zon Biome within Brazil’s borders, excluding the munici-palities pertaining to the Brazilian Legal Amazon that arenot part of the Amazon Biome.

Data and Methodology

Health outcomesBrazil has a universal health system (Sistema Único deSaúde, SUS, as its acronym inPortuguese), whichprovidesfree health care to Brazilian citizens. The SUS has an in-formatics department (DATASUS) that assembles data onhospitalizations – which includes the days a person re-mains hospitalized and the cost to the SUS- mortality andbirths at the individual level. FromDATASUS, therefore,weobtained administrative data on health outcomes.

We gathered monthly data at the municipality level fromJanuary 2010 to December 2019. There are 546municipal-ities in the Brazilian Amazon (Amazon biome). Given thesizeof the regionand thedemographicdensity, not allmu-nicipalities are equipped with hospitals. Hence, we trackthemunicipality of residence of each person thatwas hos-pitalized. Following the literatureon thee�ects of air qual-ity - related to fires – and health, we focused on: respira-tory diseases. [1] [2]

As regards the age structure of the population mostly af-fected, He et al. [2] argue that older people in China suf-fer more from ambient air pollution related to straw burn-ing. Rangel and Vogl [3] focus on infant health in a studyon the e�ects of agricultural fires at SãoPaulo state, Brazil.Jacobsen et al. [4] study the e�ects of particulate matterand black carbon from seasonal fires in schoolchildren’shealth in the Brazilian Amazon. Hence, our focus will beon infants (0-1-year-old), children between 1 and 5 yearsold and those with more than 60 years old.

∗This Technical Note is part of a contribution of the Instituto de Estu-dosparaPolíticasdeSaúde (IEPS) to the joint report by IEPS, the InstitutodePesquisaAmbiental para aAmazônia (IPAM) andHumanRightsWatch.†Visiting professor at the Universidade Federal Fluminense.

For every health outcome, we standardize data to facili-tate comparisons acrossmunicipalities and time. Becausemonthly, age-stratified population is not available, wewillcompare health outcomes in relation to monthly numberof admissions/100,000 total inhabitants per municipality.We use the same approach to age-specific groups rates ofhospital admissions. However, there is not, for the yearsin the sample, estimates of municipal population by agegroups. Therefore, we utilize the rate of hospitalization byage group divided by total population. To the extent thatthe age distribution of the population does not vary sub-stantially, this should not a�ect our estimates.

Deforestation, Fires and Ambient Air PollutionMonthly data on deforestation and active fire hotspots bymunicipalityused in thisprojectwereassembledby the In-stituto de Pesquisa Ambiental da Amazônia (IPAM), a part-ner institution in this project. IPAM obtained and analyzeddata from the Brazilian National Institute for Space Re-search (INPE) databases on deforestation and active firehotspots. The data is structured in a panel at the munic-ipality level and with a monthly structure.

The data on deforestation is based on DETER-Bwhich rep-resents periodical deforestation alerts that were aggre-gated by month from August 2016 to December 2019. Itsmain di�erence from previous DETER data is the spatialresolution that increased from 250m to 60m by addingthese other satellites to Modis daily product. In addition,DETER-B brings many classes indicating disturbances innative vegetation such asmining, logging, forest degrada-tion by fire and deforestation (forest clear cut). For thisstudy, we only considered the classes associated with de-forestation. 1 DETER-B data gives a good perspective ofthe deforestation trends since it presents daily informa-tion that can be aggregated by month or year.

The data on fires measures specifically number of activefire hotspots collected from the reference satellite used byINPE tomonitor the spatial and temporal dynamics of firesin Brazil. This data is based on the MODIS sensor on boardof the Aqua satellite. This data correlates with the areaburned and is a reliable indicator on the overtime trendsof fire activity.2 The active fire hotspots dataset representsthebestdataset to comparewithhealth issues since it cap-tures the dynamics of fires over time.

Data onmonthly ambient air pollution at themunicipalitylevel was obtained from the System of Environmental In-formation Related to Health (Sistema de Informações Am-

1The DETER data is available in the INPE website (access on Jun. 182020)

2The data is available in the INPE website (access on Jun. 18 2020)

Aug. 2020 1 of 7

Nota Técnica n. 11

bientais Integrado à Saúde, SISAM). We will mainly utilizedata on PM2.5 but will also inspect correlations with otherpollutants such as CO, SO2, NO2 and O3.3

Additional variablesIn order to understand or control for some covariates thatmight be related to hospitalization rates in the BrazilianAmazon, wewill include data on precipitation, relative hu-midity and temperatures, extracted from INPE.

Table A.1 in the appendix provides descriptive statistics oneach variable as well as their source. All variables are atthe municipality level and are monthly.

MethodologyWe will conduct a statistical analysis to assess the corre-lation between ambient air pollution from fires related todeforestation and health outcomes in the Brazilian Ama-zon. Ideally, wewould like to recover a causal estimate, asour estimates will control for some observable variables,as temperature, rainfall and moisture, and unobservablevariables specific to each municipality that do not vary intime. This design eliminates confounding from season-ality and climate shocks such as El Niño years. However,even a�er conditioning our results on those variables, de-termining causality is challenging due to omitted variablebias. One likely confounder is the association betweenfires and economic activity, whichmight a�ect health out-comes through other channels. As there is not availablemonthly economic data at themunicipal level, wewill addspecific municipality trends to account for specific trendsthatmightbe related toeconomicactivity aswell ashealthpolicies at the municipal level.4

We hypothesize that ambient air pollution due to fire ac-tivity in a given municipality might lead to an increase inhospitalizations andmortality due to respiratory illnesses.We address this correlation by relying on the panel struc-ture of the data, which allows us to control for unobserv-able variables from municipalities and common monthlyshocks, by includingmunicipality andmonth by year fixede�ects. Therefore, our benchmark model to be estimatedis:

Ymt = β0 +β1 ∗AirPol lut ionmt + γ ∗ Xmt +λm +αt + εmt

Where Ymt is a variable that measures health output. Inour case, we will use alternative health outcomes, such ashospitalizations andmortality due to respiratory diseasesby age groups. AirPol lut ionmt is a variable thatmeasures

3http://queimadas.dgi.inpe.br/queimadas/sisam/v2/dados/download/

4Another possible omitted variable that might lead to inconsistentestimator is indoor air pollution from cooking fires. However, we be-lieve that the inclusion of municipality fixed e�ects as well as munici-pality trends might deal with this caveat. Moreover, as Gioda et al. [5],report, cooking with firewood represents a relatively small fraction, be-tween 3.2% and 4.5% of households in Brazil.

the concentration of a given pollutant (PM2,5, CO, NO2 andSO2) in the municipality m during month t. O The coe�i-cient - β1 - is our coe�icient of interest and recovers ourstudied correlation conditional on our control variables.Xmt is a vector of covariates that might also a�ect healthoutcomes. Control variables are mean rainfall, mean tem-perature and mean humidity. The term αt is a time fixede�ect, which captures month by year shocks common toall municipalities, whereas λm is the municipality fixed ef-fect, which captures e�ects of unobservable and invariantvariables in time. The model error term is εmt .

Trends in respiratory health in theBrazilian AmazonAs regards respiratory diseases, hospitalizations in theBrazilian Amazon biome follow a seasonal pattern, whichreach a maximum around May – which marks the begin-ning of the driest season in the Amazon. Figure 1 plots hos-pitalizations related to respiratory diseases from January2010 to December 2019.

Figure 1. Hospitalizations due to respiratory diseasesin the Amazon biome

Source: DATASUS

Overall, therewasadeclining trend inhospitalizationsdueto respiratory diseases in the Amazon. It seems, however,that this trend has achieved a certain stability from 2015and there is a clear seasonal peak around May.

In the Brazilian Amazon, the concentration of pollutants ismainly related to fires from deforestation [6]. This is whywe are able to observe a peak in the atmospheric concen-tration of pollutants, as PM2.5 fromAugust to October. Fig-ure 2 displays the pattern of PM2.5 concentration. The hor-izontal line in Figure 2 represents the Air Quality Guide-line threshold established by the World Health Organiza-tion for 24hmean.

On Figure 3, we observe the trends, from 2016 to 2019, offocusof active firehotspots in theBrazilianAmazonForest.The period of active fire counts reflects the same period ofdata from DETER-B. As it is clear, there is a seasonal pat-

Aug. 2020 2 of 7

Nota Técnica n. 11

Figure 2. Mean of PM2.5 concentration in the Amazonbiome

Source: INPE

ternwhere fires start around July and last until Septembereach year.

Figure 3. Active fire hotspots in the Brazilian AmazonBiome

Source: IPAM, based on data from INPE/BDqueimadas

In addition toactive firehotspots,wemakeavisual inspec-tion of the trends regarding deforestation. As shown onFigure 4, there is a clear peak in deforestation from July2019 to September 2019 that does not have a parallel inprevious years. The surge in deforestation last year helpsin understanding the extent of the problems associated tosmoke from fires last year.

ResultsAmbient Air Pollution, Fires and DeforestationThe majority of deforested plots in the Brazilian Amazonis burned to clean land, in preparation for cattle ranching,agriculture and mining activities [7]. The burned biomassgenerates many pollutants, as CO2, CO, NOx and particu-late matter (especially PM2.5) [8]. As Barcellos et al. [9] ar-gue, particulatemattermay travel hundredsof kilometers,

Figure 4. Deforestation (km2/month) in the BrazilianAmazon Biome

Source: INPE

spreading pollution even to regions not a�ected by defor-estation.

Ultimately, the health impacts of fires related to deforesta-tion in the Amazon can be sensed distant from the firesources. Therefore, when assessing health impacts, themain independent variable to be utilized in this report isthe concentration of particulate matter, which has signif-icant impacts in human health [10]. The e�ects of vege-tation fire events are so expressive that they have evendeserved a guideline document produced by the WorldHealth Organization [11].

Table A.2, in the annex, displays the relationship betweenthe concentration of di�erent pollutants and fire and de-forestation activities. All pollutants have a positive corre-lation between their concentration in the atmosphere andfires related to deforestation.

Health and Ambient Air PollutionHospitalization and overall populationTable 1 displays the relationship between hospitalizationrates due to respiratory diseases and themonthlymean offine particulate matter (PM2.5µg/m3). From Columns (1)-(3), the dependent variable is the overall hospitalizationrate (per 100,000 inhabitants) due to respiratory illness. Incolumn (4), we assess the e�ects on the number of hos-pitalization days due to the same kind of diseases. Fromcolumns (1) to (3), we estimate the e�ect of the level ofPM2.55 concentration on hospitalization rates, by control-ling for municipality andmonth-by-year fixed e�ects (Col-umn (1)); weather covariates (Column (2)) andamunicipal-ity trend (Column (3)). We add these specific municipalitytrends to account for the consistent reduction in hospital-ization in time, as seen on Figure 1.

Results point to a positive relationship between ambientair pollution and hospitalization rates. From column (3),an increase of one standard deviation (17.88) in pollutionlevels relates to an increaseof 0.82 inhospitalization rates,

Aug. 2020 3 of 7

Nota Técnica n. 11

Table 1. Hospitalization rates due to respiratory illness and ambient air pollution

(1) (2) (3) (4)VARIABLES Hosp. Rate Resp. Hosp. Rate Resp. Hosp. Rate Resp. #Days of Hosp.

PM2.5 0.026* 0.053*** 0.046*** 0.070**(0.013) (0.013) (0.012) (0.033)

Observations 65,520 65,520 65,520 65,520R-squared 0.502 0.503 0.598 0.931Municipality FE Y Y Y YMonth by Year FE Y Y Y YWeather variables N Y Y YMunicipality Trend N N Y YNumber of municipalities 546 546 546 546Mean of Dep. Var. 61.90 61.90 61.90 106.9

Notes: Table reports OLS estimates with municipality and month-by-year fixed e�ects. In columns (1)-(3), de-pendent variable is the rate of hospitalization (per 100,000 inhabitants) due to respiratory illness. In column(4), dependent variable is the overall number of monthly hospitalization days due to respiratory illness. Ro-bust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Source: own elaboration.

or 1.33% of the mean of the dependent variable. A onestandard deviation in PM2.5 concentration is equal to anincrease from the total average (16.42) to the average inSeptember, the month where pollution levels are at theirhighest (34.33). As regards the number of hospitalizationdays, an increase of one standard deviation in our mea-sure of ambient air pollution relates to 1.25 more days inhospital, or 1.17%of the averagenumber of hospitalizationdays.

On Table 1, we have focused on particulate matter. As ar-gued by Chan [12]), health risks related to particulatemat-ter (both PM10 and PM2,5) are especially well documented.This is so because fine particulate matter can penetratedeep into the lung. However, other pollutants also imposea risk on health [13]. Hence, on Table A.3 of the Annex,we focus on CO, NO2 and SO2 and their respective rela-tionshipswith respiratory diseases. We estimate the samespecification as in Column (3) from Table 1, with munici-pality trends.

An importantpoint raisedbyGra�-ZivinandNeidell [14] ) isthat pollution can have nonlinear e�ects on health. Thus,we allow for nonlinear e�ects by examining how hospi-talization rates vary according to dummies related to theWorld Health Organization Air Quality Guidelines thresh-olds [15]. We follow the procedure proposed by Currie etal. [16]), who used instead the U.S. National Air QualityStandards. Figure A.1 in the appendix displays the esti-mated coe�icients for the thresholds proposed by WHO[15].The results show a non-linear relationship, where it isonly above interim target 1 that we see a statistically sig-nificant relationship between air pollution and hospital-izations, which confirms the nonlinear e�ects found in theliterature. This result means that when ambient air pol-lution is above 75 µg/m3 in 24-hour mean, the hospital-ization rate due to respiratory diseases is 8.7 per 100,000people higher as compared to WHO air quality guidelinefor PM2.5. This can also be interpreted as an increase of14.8% as compared to average hospitalization rate in 2019

due to respiratory diseases.

Hospitalization and Health by Age GroupsWe utilize the rate of hospitalization by age group dividedby total population (Hosp. by Age Group/100,000 totalpopulation). Table 2 provides estimates of hospitalizationdue to respiratory diseases by age groups. The age groupon column (1) comprises children with less than one yearold. Column (2) displays estimates of hospitalization inchildren with more than one year old and less than sixyears old.5 On column (3), the age group is of children be-tween six and fourteen years old.6 Finally, on column (4),weprovideestimates forhospitalizationamong thosewithmore than 60 years old.7

Results from Table 2 follow as expected, given the litera-ture review. Both children and older people su�er withrespiratory diseases as ambient air pollution increases.However, among children, the e�ects are only felt amongthose with less than one year old. Using the estimated co-e�icient from column (1), a one standard deviation in am-bient air pollution is related to an increase 1.5% in meanhospitalization rates to infants. The same evaluation forthe elderly implies a related increase of 3.0% in hospital-ization rates.

It seems that the extremes of the age groups aremore sus-ceptible to the e�ects of ambient air pollution on health.However, this correlation appears to be even strongeramong older people, where pollution peaks can have im-portant morbidity e�ects.

5Rodrigues et al. [17] analyze the distribution of hospitalization forrespiratory diseases in children with less than 5 years old in the state ofRondônia. Carmo et al. [18] also focus on children with less than 5 yearsold in Alta Floresta, MT.

6Jacobson et al. [4] describe the e�ects of air pollution from fires inthe Brazilian Amazon on peak expiratory flow on children between 6 to15 years old living in the municipality of Tangará da Serra, MT.

7Carmoet al. [18] discuss the respiratoryproblems related to fires andparticulate matter among people with more than 64 years old.

Aug. 2020 4 of 7

Nota Técnica n. 11

Table 2. Hospitalization rates due to respiratory illness and ambient air pollution

(1) (2) (3) (4)VARIABLES [0y-1y] (1y-5y] [6y-14y] >60y

PM2.5 0.011** -0.002 -0.001 0.023***(0.004) (0.003) (0.003) (0.005)

Observations 64,296 64,296 64,296 64,296R-squared 0.397 0.433 0.412 0.397Municipality FE Y Y Y YMonth by Year FE Y Y Y YWeather variables Y Y Y YMunicipality Trend Y Y Y YNumber of municipalities 546 546 546 546Mean of Dep. Var. 13.37 9.917 6.915 13.72

Notes: Table reports OLS estimates with municipality, month-by-yearfixed e�ects and municipality trends. Dependent variable is the rateof hospitalization (per 100,000 inhabitants) due to respiratory illness.Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1.Source: own elaboration.

Mortality and overall populationA�er analyzing morbidity, in this subsection, we estimatethe correlation between mortality rates and ambient airpollution. Mortality rates are represented as the numberof deaths related to a specific source divided by total pop-ulation (deaths per 100,000 inhabitants). Table 3 displaysthe results for deaths related to respiratory diseases.

Overall, it seems that there is not, conditional on fixed ef-fects, a correlation in the short term between mortalitydue to respiratory conditions and ambient air pollutionin the Brazilian Amazon.8 These results seem to be gen-eral: as we inspect the relationship among distinct agegroups, the estimated coe�icient cannot be distinguishedfrom zero in any age group, as shown on Table A.4, in theannex.

Discussion of resultsThis report has analyzed morbidity and mortality relatedto ambient air pollution in the Brazilian Amazon, definedas the Amazon Biome in Brazil. Our results show a statisti-cally significant relationshipbetweenhospitalizationsdueto respiratory diseases and the concentration of particu-latematter. Moving forward, it is important for policymak-ers to understand the scope of the public health impact inthe Amazon region and the strain it places on its precari-ous health infrastructure, with a view to formulate preven-tive and proactive protective measures for a�ected com-munities and, particularly, vulnerable groups. Thus, natu-ral questions arise such as how many people were hospi-talized in 2019 due to the decrease in air quality related tothe outbreak of fires and the extent of deforestation in theBrazilian Amazon? Given that, what was the fiscal cost tothe public health system? Which age groups were mainly

8This result should be viewed with since we are investigating short-term correlations. For an analysis of the e�ects of biomass burning onDNA damage and human lung cells, see De Oliveira Alves et al. [19]

a�ected?

In this section, we will address these questions, based ontheestimatespresented in theprevious section. Firstly,wetake the predicted values of ambient air pollution, basedon the estimates of Table 1, Column (3). Then, we applythese values to assesshowmanypeoplewerehospitalizedin 2019 linked to the decrease in air quality related to de-forestation and fires, based on estimates from Table 1. Inother words, we compare the number of hospital admis-sions in 2019 related to air pollution from fires and defor-estation with a counterfactual of zero deforestation andzero fires. We aggregate the results and use each munic-ipality population to have overall figures for the numberof total hospitalizations in 2019 related to the outbreak offires and deforestation.

In 2019, we estimate that there was a total of 2,195 hospi-talizations (with a 95% confidence interval between 1,098and 3,291) due to respiratory diseases linkedwith ambientair pollution from deforestation-related fires. This led toa total of 6,698 days of hospitalization (with a 95% con-fidence interval between 1,292 and 12,105 days). In 2019,theaveragecost for hospitalizationsdue to respiratorydis-eases in the Brazilian Amazon was R$ 858.15 (USD 217.50)per hospitalization which we apply to the 2,195 hospital-izations. The amount of reimbursements made by thefederal government was of R$ 1.88 million (USD 477,551).However, as highlighted by Américo and Rocha [20], stateand municipal governments also add funds on top of fed-eral transfers, in a ratio that proxies 2:1. Hence, applyingthis ratio, we estimate that total costs associated to hospi-talizations due to deforestation-related fires was R$ 5.64million (USD 1.4 million).

For the same year, during the period of June-Octoberwhich encompasses the dry season in the Amazon [21],we estimate, that there was a total of 1,012 hospitaliza-tions (with a 95% confidence interval between 506 and

Aug. 2020 5 of 7

Nota Técnica n. 11

Table 3. Mortality rates due to respiratory illness and ambient air pollution

(1) (2) (3)VARIABLES Mort. Rate Resp. Mort. Rate Resp. Mort. Rate Resp.

PM2.5 0.000 0.000 -0.000(0.001) (0.001) (0.001)

Observations 58,944 58,944 58,944R-squared 0.066 0.066 0.081Municipality FE Y Y YMonth by Year FE Y Y YWeather variables N Y YMunicipality Trend N N YNumber of municipalities 546 546 546Mean of Dep. Var. 2.844 2.844 2.844

Notes: Table reports OLS estimates withmunicipality andmonth-by-year fixed e�ects. De-pendent variable is the rateofmortality (per 100,000 inhabitants) due to respiratory illness.Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Source: own elabo-ration.

1,517) due to respiratory diseases associatedwith ambientair pollution from deforestation-related fires. This led toa total of 3,021 days of hospitalization (with a 95% confi-dence interval between 583 and 5,460 days). The cost tothe public healthcare system (SUS) between June and Oc-tober 2019 resulting from hospitalizations due to respira-tory diseases linked to deforestation-related fires was ofR$ 2.6 million (USD 660,000).

The remarkable spike in fire hotspots in August 2019 wasaccompanied by an increase in excess hospitalizationslinked to ambient air pollution. Hospitalizations due torespiratory diseases linked to the increase in ambientair pollution from deforestation-related fires increased by65% between July and August, the month when the firespeaked. The burden of ambient air pollution among agegroups shows that infants and older peoplewere themosta�ected. In 2019, there were 467 (with a 95% confidenceinterval between 109 and 824) hospitalizations of infantsaged 0-1-year-old and 1,080 (with a 95% confidence in-terval between 608 and 1,552) hospitalizations of olderpeople aged 60 years old and over due to respiratory dis-eases linked to the increase in ambient air pollution fromdeforestation-related fires. For the period concerning onlythe dry season, between June andOctober, therewere 215hospitalizations of infants and 47 older persons. The sumof these two age groups amounts to 70% of total hospital-izations due to respiratory diseases linked to air pollutionfrom deforestation-related fires in 2019.

As a conclusion, our estimates showan important e�ect oftheoutbreakof fires anddeforestation in 2019onhospital-izations due to respiratory diseases. In 2020, with higherrates of deforestation and, relatedly, a higher risk of fires,as well as the devastating toll of COVID-19 in the Amazonregion, there is a justifiable fear of a serious threat to thepopulation in the Brazilian Amazon and its strained publichealth system.

References[1] C. L. Reddington, E. W. Butt, D. A. Ridley, P. Artaxo, W. T. Mor-

gan, H. Coe, and D. V. Spracklen. Air quality and humanhealth improvements from reductions in deforestation-related fire in brazil. NatureGeoscience, 8(10):768–771, 2015.

[2] Guojun He, Tong Liu, and Maigeng Zhou. Straw burning,pm2.5, and death: Evidence from china. Journal of Devel-opment Economics, 145:102468, 2020.

[3] Marcos A. Rangel and Tom S. Vogl. Agricultural fires andhealth at birth. The Review of Economics and Statistics,101(4):616–630, 2019.

[4] Ludmilla da Silva Viana Jacobson, Sandra de Souza Hacon,Hermano Albuquerque de Castro, Eliane Ignotti, Paulo Ar-taxo, Paulo Hilário Nascimento Saldiva, and Antonio CarlosMonteiro Ponce de Leon. Acute e�ects of particulate mat-ter and black carbon from seasonal fires on peak expiratoryflow of schoolchildren in the brazilian amazon. PLOS ONE,9(8):1–14, 08 2014.

[5] Adriana Gioda, Gisele Birman Tonietto, and AntonioPonce de Leon. Exposição ao uso da lenha para cocção nobrasil e sua relação com os agravos à saúde da população.Ciência Saúde Coletiva, 24:3079 – 3088, 08 2019.

[6] C. L. Reddington, W. T. Morgan, E. Darbyshire, J. Brito,H. Coe, P. Artaxo, C. E. Scott, J. Marsham, and D. V.Spracklen. Biomassburning aerosol over the amazon: anal-ysis of aircra�, surface and satellite observations using aglobal aerosol model. Atmospheric Chemistry and Physics,19(14):9125–9152, 2019.

[7] Ronaldo Seroa da Motta, Mário Jorge Cardoso Mendonça,Daniel Nepstad, Maria del Carmen Vera Diaz, Ane Alencar,João Carlos Gomes, and Ramon Arigoni Ortiz. O custoeconômico do fogo na amazônia.

[8] Saulo R. Freitas, Karla M. Longo, M. A. F. Silva Dias, andP. L. Silva Dias. Emissões de queimadas em ecossistemasda américa do sul. Estudos Avançados, 19:167–185, 04 2005.

[9] Christovam Barcellos, Diego Xavier, Sandra Hacon, PauloArtaxo, Renata Gracie, Monica Magalhães, Vanderlei Matos,

Aug. 2020 6 of 7

Nota Técnica n. 11

Table 4. Estimates for absolute hospitalizations due to respiratory diseases linked to increased ambient air pollu-tion from deforestation-related fires in the Brazilian Amazon Biome

Month(year of 2019)

Total number ofexcess

hospitalizationsattributable todeforestationrelated fires

Average Totalnumber of

hospitalizations(2016-2018)

Total number ofhospitalizationsof infants (0-1year old)

Total number ofhospitalizationsof older people(60 years and

older)

January 148 9,311 31 75

February 143 9,646 30 76

March 154 13,652 33 82

April 148 14,650 32 82

May 122 15,120 26 67

June 109 13,465 23 59

July 141 11,335 30 67

August 232 10,314 49 103

September 288 9,898 61 134

October 242 10,202 52 113

November 255 9,498 54 119

December 213 8,744 45 104

TOTAL 2,195 135,835 467 1,080

Antônio Miguel Monteiro, and Patrícia Feitosa. Queimadasna amazônia e seus impactos na saúde: A incidência dedoenças respiratórias no sul da amazônia aumentou signi-ficativamente nos últimos meses.

[10] Ki-Hyun KIM, Ehsanul KABIR, and Shamin KABIR. A reviewon the human health impact of airborne particulatematter.Environment international, 74:136–143, 2015.

[11] Dietrich H. Schwela, Johann G. Goldammer, Lidia H.Morawska, and Orman Simpson. Health guidelines for veg-etation fire events: Guideline document.

[12] E. Chan. Essentials for Health Protection: Four Key Compo-nents. Oxford University Press, 2020.

[13] World Health Organization. Ambient air pollution: Healthimpacts, 06 2020.

[14] Joshua Gra� Zivin and Matthew Neidell. Environment,health, and human capital. Journal of Economic Literature,51(3):689–730, September 2013.

[15] World Health Organization. Regional O�ice for Europe. Airquality guidelines global update 2005 : particulate matter,ozone, nitrogen dioxide and sulfur dioxide, 2006.

[16] Janet Currie, Eric A Hanushek, E. Megan Kahn, MatthewNeidell, and Steven G Rivkin. Does pollution increaseschool absences? The Review of Economics and Statistics,91(4):682–694, 2009.

[17] Poliany Cristiny de Oliveira Rodrigues, Eliane Ignotti, andSandra de Souza Hacon. Distribuição espaço-temporaldas queimadas e internações por doenças respiratórias em

menores de cinco anos de idade em rondônia, 2001 a 2010.Epidemiologia e Serviços de Saúde, 22:455–464, 09 2013.

[18] Longo KM Freitas S Ignotti E Ponce de Leon A et al.Carmo CN, Hacon S. Associação entre material particu-lado de queimadas e doenças respiratórias na região sulda amazônia brasileira. Revista Panamericana de SaludPública, 27:10–6, 2010.

[19] Vessoni A. T. Quinet A. Fortunato R. S. Kajitani G. S. PeixotoM. S. Hacon S. S. Artaxo P. Saldiva P. Menck C. Batistuzzode Medeiros S. R. de Oliveira Alves, N. Health transition inbrazil: regional variations and divergence/convergence inmortality. Scientific reports, 7:10937, 2017.

[20] Pedro Américo and Rudi Rocha. Subsidizing access to pre-scription drugs and health outcomes: The case of diabetes.Journal of Health Economics, 72:102347, 2020.

[21] Jose A. Marengo, Javier Tomasella, Lincoln M. Alves, Wag-ner R. Soares, and Daniel A. Rodriguez. The drought of 2010in the context of historical droughts in the amazon region.Geophysical Research Letters, 38(12), 2011.

Instituto de Estudos para Políticas de Saúde

Sant’Anna, A. A.; Rocha, R. (2020). Health Impacts ofDeforestation-Related Fires in the Brazilian Amazon. Tech-nical Note n.11. IEPS: São Paulo.

www.ieps.org.br | +55 11 4550-2556 | contato@ieps.org.br

Aug. 2020 7 of 7

Annex

Table A.1 – Descriptive Statistics

Variable Mean Std. Dev Min Max Obs Period Source

Hosp. rate – resp. (per

100,000 people) 61.90 68.12 0 1471.47 65,520 Jan 2010 – Dec 2019 DATASUS

Mort. rate – resp. (per

100,000 people) 2.84 5.19 0 96.18 58,944 Jan/2010 – Dec/2018 DATASUS

PM2.5 (g/m3) 16.38 17.88 2.6 657.71 65,520 Jan/2010 – Dec/2019 INPE

Active fire hotspots 14.07 70.26 0 35577 24,336 Jan/2016 – Dec/2019 INPE

Deforestation (in km2) 0.89 4.56 0 178.64 22,386 Aug/2016-Dec/2019 INPE

Table A.2 – Correlation between active fire hotspots, deforestation and ambient air pollution

(1) (2) (3) (4)

VARIABLES Ln_PM2.5 Ln_CO Ln_SO2 Ln_NO2

Ln(active fire hotspots) 0.151*** 0.126*** 0.064*** 0.109***

(0.007) (0.005) (0.004) (0.005) Ln(Deforestation) 0.022*** 0.014*** 0.005*** 0.010***

(0.003) (0.002) (0.001) (0.002) Rainfall 0.032*** 0.027*** 0.005*** 0.012***

(0.002) (0.001) (0.001) (0.001)

Temperature 0.104*** 0.113*** 0.098*** 0.100*** (0.007) (0.007) (0.007) (0.008)

Observations 22,386 15,834 15,834 15,834

R-squared 0.591 0.781 0.744 0.677 Municipality FE Y Y Y Y Month by Year FE Y Y Y Y Municipality Trend Y Y Y Y Number of

municipalities

546 546 546 546

Mean 2.704 4.975 0.306 0.431 Notes: Table reports Ordinary Least Squares estimates with municipality, month-by-year fixed effects and

municipality trends. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Source: own

elaboration.

Table A.3 – Hospitalization rates due to respiratory illness and other pollutants

(1) (2) (4) VARIABLES Hosp. Rate Resp. Hosp. Rate Resp. Hosp. Rate Resp.

CO 0.008***

(0.002) NO2 1.088***

(0.235) SO2 1.555*** (0.377)

Observations 58,944 58,944 58,940

R-squared 0.609 0.609 0.609

Municipality FE Y Y Y Month by Year FE Y Y Y Weather variables Y Y Y

Municipality Trend Y Y Y Number of municipalities 546 546 546

Mean 62.83 62.83 62.83 Notes: Table reports OLS estimates with municipality, month-by-year fixed effects and municipality trends.

Dependent variable is the rate of hospitalization (per 100,000 inhabitants) due to respiratory illness. Robust

standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1

Table A.4 – Mortality rates due to respiratory illness and ambient air pollution, among age

groups

(1) (3) (5) VARIABLES [0-1yr) [1-5yr] >60yr

PM2.5 -0.000 -0.000 -0.001

(0.000) (0.000) (0.001)

Observations 58,728 56,568 58,944 R-squared 0.066 0.032 0.081

Municipality FE Y Y Y Month by Year FE Y Y Y

Weather variables Y Y Y

Municipality Trend Y Y Y Number of municipalities 544 524 546 Mean 0.174 0.0924 2.016

Notes: Table reports OLS estimates with municipality and month-by-year fixed effects. Dependent variable

is the rate of mortality (per 100,000 inhabitants) due to respiratory illness. Robust standard errors in

parentheses. *** p<0.01, ** p<0.05, * p<0.1. Source: own elaboration.

Figure A.1 – Estimated coefficients for the World Health Organization’s Air Quality

Guidelines (2006) thresholds for PM 2.5 concentration in 24-hour mean

Source: own elaboration.

The estimated parameters must be interpreted as a difference to benchmark pollution, which is

WHO air quality guidelines and refers to PM2.5 less than 25 g/m3 in 24-hour mean. WHO IT 3 refers

to the Interim Target 3, which is between 25 and 37.5 g/m3 in 24-hour mean. WHO IT 2 refers to

the Interim Target 2, which is between 37.5 and 50 g/ m3 in 24-hour mean. WHO IT 1 refers to the

Interim Target 1, which is between 50 and 75 g/m3 in 24-hour mean. Finally, Above WHO IT 1

refers to air pollution above interim target 1 and is, therefore, a significant amount of air pollution.

On Figure 5, we plot the values of the estimated coefficients having the WHO Air Quality Guidelines

(WHO ACG) as a benchmark. In other words, the estimated coefficients are compared to a case

where the parameter of WHO ACG is set equal to zero.

-50

510

15

WHO IT 3 WHO IT 2 WHO IT 1 Above WHO IT 1