Assessing the spatial influence of deforestation on malaria incidence in Pará State, Amazon region, Brazil, 2008-2019

Malaria is a disease caused by parasites of the Plasmodium genus, which are transmitted by anopheline mosquitoes. Fivespecies of this genus can infect humans, with Plasmodium falciparum and P.vivax being the most prevalent.^{(1, 2, 3)} Geographically, P.vivax predominantly affects populations in various regions, including Asia,Oceania, and South America. This includes a substantial area in northernBrazil, the Amazon region.^{(4, 5, 6, 6)}

Malariatransmission is primarily confined to tropical zones, where environmentalfactors influence parasite transmission. Notably, deforestation holdsparticular significance because of its increase in tropical areas, specificallyin the Brazilian Amazon region.^{(7, 8, 9)} As a result, ecologicalfactors play a role in regulating the species composition of mosquitopopulations, influencing both the number and types of malaria vectors.^{(10, 11)}
The interplay between malaria transmission, forest cover, and deforestation isintricate, directly affecting all transmission components, namely the vector,host, and the environment.⁽⁸⁾

The incidence ofmalaria is contingent upon environmental factors that facilitate theproliferation of vector mosquitoes, including their adaptation to climate,altitude, and vegetation.^{(9, 11, 12)} Alterations to the environmentresulting from economic activities such as land use^{(13, 14)} cancontribute to the proliferation of vectors, human exposure to infectedmosquitoes, mosquito biting rates, and ultimately, the incidence of malaria.^{(14, 15)} Malaria transmission is further influenced by the living conditions of thepopulation, exhibiting a strong association with poverty and economicactivities that contribute to deforestation, such as agriculture and mining,which are common in the Brazilian Amazon.^{(16, 17, 18, 19)} Previousstudies have indicated that deforestation may increase the incidence ofmalaria.^{(8, 9, 11, 12)} Nevertheless, other reports emphasise that theimpact of this factor is contingent upon the time elapsed since the loss offorest cover,^{(20, 21)} and the rate of deforestation also influencesvector efficiency.^{(21, 22)}

In Brazil, theAmazon region accounts for 99% of malaria cases, with approximately 85%attributed to P. vivax.^{(4, 5, 14)} The highest malaria burden isobserved in rural areas,^{(14, 23)} where the past three decades havewitnessed a notable increase in deforestation. In Pará, one of the nine stateswithin the Brazilian Amazon, forest devastation is widespread, driven byeconomic activities such as settlement formation for subsistence farming, theconstruction of roads and hydroelectric dams, and large-scale deforestation forcattle ranching.^{(14, 17, 23, 24)} Another significant contributor toenvironmental changes impacting the malaria incidence is mining.^{(17, 24)} Within the Amazon region, malaria is concentrated in select municipalities andcategorised by annual parasite incidence (API), ranging from very low to highrisk.⁽²⁵⁾

This retrospective analysis explores thespatial and temporal patterns of malaria incidence alongside deforestation andits resultant landscape components (composition and configuration) over a12-year period across all municipalities in Pará State, Brazil. Theinvestigates the potential relationship between environmental changes,particularly deforestation, and malaria incidence in this endemic region.

MATERIALS AND METHODS

Study area and data collection - This study included all 144 municipalities in Pará State, wherethe annual malaria incidence rates during the study period (2008-2019) rangedfrom 0 to 750 malaria cases per 1,000 people in the Brazilian Amazon region(Fig. 1). These data were retrieved for each municipality of infection fromofficial malaria data repositories held by the Brazilian Ministry of Health’smalaria surveillance system (SIVEP-Malaria) under the Brazilian Malaria ControlProgramme.⁽²⁶⁾ API, defined as the total number of new malaria casesdivided by the total number of examinations per 1,000 people in a given year,was retrieved from SIVEP-Malaria. To evaluate temporal and spatial differencesin malaria rates, these API values were averaged across three 4-year periods:2008-2011, 2012-2015, and 2016-2019. Approval for this study was obtained fromthe Ethics Committee at the Federal University of Pará (approval number5.137.483).

Deforestation metrics were calculated usingpublicly available data obtained from MapBiomas v. 9.⁽²⁷⁾ TheMapBiomas collection, derived from Landsat imagery with a 30-metre spatialresolution, provides annual land use and land cover maps spanning 38 years(1985-2023) (Fig. 2). Deforestation was calculated as loss of forest cover (%)in the previous year, the proportion of forest converted to pasture (%), theproportion converted to urban areas (%), the proportion of retained/restoredforest (%), and the number of remaining forest patches (fragmentation). Waterlevels (%) were included as a control variable. These variables were averagedacross the same temporal periods (2008-2011, 2012-2015, and 2016-2019) toexplore their association with malaria incidence rates.

Socioeconomic and climate data were obtainedfrom the Fundação Amazônia de Amparo a Estudos e Pesquisas do Pará to supportthe analysis of the relationship between malaria incidence and deforestationlevels.⁽²⁸⁾ Poverty and extreme poverty, as defined by the UnitedNations — poverty being the deprivation of basic human needs and extremepoverty reflecting the inability to meet even basic survival conditions — weremapped using the regions of integration, the official administrative divisionsestablished by the state government of Pará (Fig. 3). Additionally, the HumanDevelopment Index for Pará State increased from 0.666 in 2012 to 0.704 in 2019,reflecting positive progress. However, significant challenges remain inensuring universal healthcare, access to higher education, and equitable incomedistribution for the entire population. Finally, all municipalities are locatedwithin a climatic zone classified as humid equatorial, characterised by twodistinct seasons: nine months of humidity and three months of dryness. Theaverage temperature ranges from 25ºC to 27ºC.⁽²⁸⁾

Spatial analysis - A spatial analysis of therelationship between malaria incidence rates and deforestation metrics wasconducted for each temporal period (2008-2011, 2012-2015, and 2016-2019) usingArcGIS for Desktop (v.10.4.1) with the Spatial Analyst extension (Esri@ArcMap™, Redlands, CA, USA). A projected coordinate system, WGS 1984 UTM Zone22S, was used for all analyses. A spatial weights matrix was constructed usinga squared inverse Euclidean distance of 250 km. This distance was selectedbased on the large territorial size of certain municipalities, such as Altamiraor São Félix do Xingu, which has an area comparable to that of several Europeancountries. The spatial weights matrix was employed to capture the spatialstructure and connectivity among municipalities in the state. This matrixaccounted for territorial area in km² and spatial connectivity (withan average of 40 neighbours, ranging from 1 to 74) and was used to adjust therelationship between malaria incidence rates and deforestation metrics.

First, anexploratory regression of malaria incidence rates for each temporal period wasperformed against deforestation metrics to test for residual normality andspatial stationarity using the Jarque-Bera and Koenker statistics,multicollinearity through the variance inflation factor, and spatialautocorrelation with the Global Moran Index. These tests informed the selectionof the appropriate spatial regression model. Based on the results, ageographically weighted regression (GWR) approach was applied to provide a morenuanced assessment of the relationship between malaria incidence anddeforestation across a heterogeneous landscape characterised by varying levelsof deforestation and its associated outcomes.

In the GWR approach, each municipality and itsneighbouring areas were modelled using local regression analyses for malariaincidence and deforestation metrics, allowing for individual interpretation ofthe results. As an initial criterion, the standard residuals for each model,ranging from -1.5 to 1.5, were used to assess how well the model fit the actualdata and to select the most accurate models for further analysis. In the Rprogramming environment (v. 4.3), the number of significant coefficients for eachdeforestation metric and temporal period was calculated. The significance ofeach coefficient was determined by dividing its estimated value by its standarderror to obtain a t-value. Coefficients with a t-value of ≥ 1.96 wereconsidered significant (i.e., p < 0.05). The municipalities wheresignificant coefficients representing the relationship between malariaincidence and deforestation metrics were found were mapped to facilitate theinterpretation of these results.

RESULTS

The deforested area in the Amazon region ofPará State over a 12-year period totalled 30,000 km², accounting forapproximately 2.5% of the state’s total area of 1,248,000 km² (Fig.2). The eastern region of Pará experienced the highest deforestation, whichextended significantly toward the south. Of particular note is the central partof the state, where a large arch-shaped area, connecting the east to the west,is commonly referred to as the arch of deforestation. In the western region,deforestation is primarily concentrated in two micro-regions: the Lower Amazonand the Tapajós River. While deforestation is widespread throughout Pará, it isparticularly concentrated along riverbanks.

The incidence of malaria cases was widespreadacross nearly all municipalities in the state, but the highest rates wereconcentrated in a small number of municipalities (Fig. 1). During this period,679,846 malaria cases were recorded, including 566,014 cases of P. vivax (83%), 94,564 cases of P. falciparum (14%), 10,319 mixed infections(2%), and 8,949 cases caused by other malaria parasites (1%). Municipalitieswith malaria incidence rates exceeding 50 cases per 1,000 people were clusteredin the regions of integration of Marajó (including Afuá, Anajás, Bagre, andCurralinho), Tapajós (Itaituba, Jacareacanga, and Trairão), Xingu (includingAltamira, Anapu, and Pacajá), and Tocantins (Oeiras do Pará). Thesemunicipalities can be classified as high-risk areas for malaria transmission.

The regions of integration of Marajó andTocantins had 90% of their population below the poverty line in 2012-2015 (Fig.3). During this period, these regions accounted for 90,945 malaria cases,representing 71% of the total malaria cases in the state. The Marajó regionremained predominantly below the poverty line in the 2016-2019 period andcontinued to include municipalities with high-risk malaria transmission.Poverty affected 80% of the population in the integration region of Xinguduring both periods (2012-2019), while in the Tapajós region, poverty slightlydecreased. In total, 74,249 malaria cases were reported across both regions,accounting for 11% of the total malaria cases in the state during the sameperiod.

The exploratory regression revealed that thespatial relationship between malaria incidence rates and deforestation metricsresulted in models with non-normal residuals, lack of stationarity, and spatialautocorrelation, but no evidence of multicollinearity. Based on these findings,the GWR approach was selected to account for spatial heterogeneity in the data.This method was particularly useful for modelling the relationship betweendeforestation and malaria incidence because it is expected to vary acrossspace, allowing for the analysis of localised variations in the estimation ofcoefficients. It provided a clearer understanding of how these variablesinteracted across different regions. For instance, the fitted GWR models foreach municipality showed varying levels of model accuracy across the state(Fig. 4). The highest accuracy was observed in 138 models for both the2008-2011 and 2012-2015 periods, and in 133 models for 2016-2019.

The GWR models explained 31%, 29%, and 46% ofthe variance in malaria incidence rates across the three temporal periods,respectively (Table). The analysis identified 12 models with significantcoefficients during 2008-2011, all indicating a non-reciprocal (negative)relationship between malaria incidence rates and municipalities with largerareas of pastureland. Additionally, two models from this period revealed a morecomplex dynamic in the Tapajós region (Fig. 5), where both deforestation andforest cover exhibited negative effects. This finding suggests that in alreadydeforested areas, such as those in Itaituba and Trairão, an increase in forestcover (e.g., through restoration) was associated with a rise in malariacase numbers.

During the 2012-2015 period, 10 modelsdemonstrated significant coefficients (Table). Of these, nine indicated anegative relationship between malaria incidence rates and larger areas ofpastureland. Additionally, one model highlighted that forested areas with highfragmentation, such as those in Altamira (Fig. 5), were more likely toexperience elevated malaria incidence rates.

During the 2016-2019 period, 56 models hadsignificant coefficients (Table). Among these, 34 indicated a negativerelationship between malaria incidence rates and pasture, consistent withfindings from earlier periods. Additionally, 55 models demonstrated a positiverelationship between forest cover and higher malaria incidence rates. Twomodels revealed more complex dynamics in the Tapajós and Xingu regions (Fig.5). In Itaituba, increased deforestation and greater forest fragmentation inareas with reduced forest cover were linked to higher malaria incidence rates.Similarly, in Altamira, forested areas undergoing recovery and exhibiting highlevels of forest fragmentation also showed elevated malaria incidence rates.

DISCUSSION

This study examined the incidence of malaria inthe Brazilian Amazon rainforest, focusing on endemic areas with varying levelsof accumulated deforestation. In the high-transmission eastern area of Pará,Anajás on Marajó Island exhibits minimal deforestation, whereas Oeiras do Paráand Pacajá have undergone extensive deforestation. In western Pará, whererecent deforestation is prevalent, Itaituba and Jacareacanga report elevatedmalaria transmission. These five municipalities, selected for their historicalsignificance, accounted for nearly 50% of Pará’s malaria cases from 2008 to2019.

Since the 1970s, following the inauguration ofthe Belém-Brasília and Transamazônica highways, significant human migrationdriven by economic opportunities has sustained high levels of bothdeforestation and malaria transmission in the Amazon region. Determinants ofdeforestation in Pará include extensive agricultural projects, cattle ranching,and the creation of large reservoirs for hydroelectric dams.^{(7, 14, 17, 23)} The GWR models presented here show that water levels in the landscape are notassociated with contemporary malaria incidence rates, despite thewell-documented explosive rise in malaria cases linked to the construction ofthe Tucuruí hydroelectric plant between 1975 and 1984. This period led to ahigh transmission risk and an outbreak of the disease upon completion of theproject.⁽⁷⁾ Additionally, the GWR models indicate that pasturelandsnow have a negative effect on malaria transmission risk, although it has beendocumented that at the onset of such activities, malaria can emerge as anendemic disease, closely linked to changes that increase exposure to malariavectors.^{(16, 21, 22, 29)} This result supports the unimodal relationshipbetween malaria risk and the landscape in the Amazon, demonstrating higher riskin areas where forests are being converted and lower risk in regions that arealready heavily degraded.^{(9, 21, 29, 30)}

The housingconditions along riverbanks in the Marajó region, particularly in Anajás, whichare associated with high levels of poverty, contribute to increased exposure tomalaria vectors. During the past two decades, Anajás has consistently beenclassified as a high-risk area for malaria.⁽²⁵⁾ Additionally, theTapajós and Xingu regions, particularly Itaituba and Altamira — some of themost remote areas in Pará — have experienced recent population growth driven bymigration, alongside local economic activities such as mining (more frequent inItaituba) and agricultural land use.^{(16, 24)} The GWR models revealthat these regions are characterised by high forest cover (> 50%) but areexperiencing forest fragmentation and ongoing modifications in forest cover,including both forest loss and restoration. Forest loss and restoration are partof a broader process of landscape alteration, driven by the conversion offorests into economically profitable land for rotational cattle ranching.^{(21, 29)} This rotation allows the land to remain productive for cattle farming whileabandoned plots regenerate, creating new habitats for malaria vectors,particularly Nyssorhynchus darlingi (formerly Anopheles darlingi),a species considered an intermediate forest disturbance specialist.^{(20, 30)} As a result, malaria risk is heightened due to these activities, emphasisingthe need for intensified surveillance efforts to reduce its incidence in theTapajós region, which has been recently impacted by deforestation.

Malaria incidence rates declined over the threestudy periods, although some municipalities exhibited variable transmissionpatterns. For instance, Oeiras do Pará had an average malaria incidence of 258per 1,000 people during 2008-2011, which decreased to 49 in 2012-2015 but roseagain to 169 in 2016-2019. This variation is common across all states of theBrazilian Amazon.^{(6, 31, 32)} These fluctuations are driven not only byeconomic activities, as discussed earlier, but also by changes in malariacontrol efforts.^{(24, 25, 33)} Malaria control measures — includingprompt diagnostics, adequate treatment, and vector control throughinsecticide-treated bed nets — were implemented in all municipalities as partof the National Malaria Control Programme, in which Pará State is a keyparticipant.^{(2, 3, 26)} However, even with these control measures,transmission rates may stabilise due to specific challenges, such as thepresence of asymptomatic reservoirs carrying P. vivax and outdoor bitingby malaria vectors such as Ny. darlingi.^{(14,32,34-37)} In asimulated scenario, if all or any of these control measures were abruptlysuspended, the incidence of malaria could increase exponentially inmunicipalities with malaria incidence rates of 10 per 1,000 people or higher.

The complexrelationship between deforestation and malaria incidence^{(8, 12, 33, 35)} complicates its interpretation, with studies showing varying results.^{(9, 11, 20, 38)} Generally, in the Amazon, initial deforestation in newly settled forest areastends to increase malaria risk up to a certain threshold of forest cover andsocial development.^{(22, 29, 36)} Beyond this threshold, asdeforestation progresses, the risk of malaria may decrease.^{(30, 33)} The GWR models provided two key insights: first, large pasturelands are notconducive to malaria transmission, even if deforestation continues; second,regions with high forest cover under pressure from forest loss andfragmentation are likely to experience higher malaria transmission. Thesefindings align with previous studies^{(20, 21, 29, 30, 33)} that emphasisethe intricate relationship between land-use changes and malaria risk.

While ongoing discussions persist, a recentinvestigation⁽³⁹⁾ identified Pará State as having the highestcumulative deforestation between 2003 and 2022. This analysis revealed thatfour of the top five most heavily deforested indigenous territories and threeof the five most impacted conservation units were located within Pará.Additionally, the study⁽³⁹⁾ found that a 1% increase indeforestation, with a 1-month lag, corresponded to a 6% rise in malaria casesat the municipal level. Pará is also home to one of the highest diversities ofmalaria vector species.⁽⁴⁰⁾ Given the vast ecological,sociodemographic, economic, and epidemiological diversity of the BrazilianAmazon, it is crucial to dissect the multifaceted drivers of deforestation andsubsequent malaria transmission. Only by understanding these dynamics canactionable insights support Brazil’s ambitious malaria elimination efforts.⁽³²⁾

This study provides valuable insights into thelinks between deforestation and malaria in Pará State using robust methods andaccessible data. However, it is important to acknowledge the limitations ofthis approach. Aggregating data over multiple years and conducting analyses atthe municipal level may obscure localised or short-term variations. While thisapproach offers a broad perspective, it may overlook finer-scale dynamics ofmalaria transmission patterns.

In conclusions -Pará State continues to face challenges in malaria elimination, with incidenceclosely linked to areas of significant forest loss and fragmentation. Thesefindings highlight how changes in forest composition influence malaria risk,emphasising the need for tailored strategies that integrate environmental andsocioeconomic factors to support Brazil’s 2035 malaria elimination plan.

ACKNOWLEDGEMENTS

To the Programa de Pós-Graduação em Biologia de Agentes Infecciosos eParasitários at the Universidade Federal do Pará.

AUTHORS’ CONTRIBUTION

CGRG, MMP, GZL and MGC conceived thestudy and conducted data acquisition; BCR, ASS Jr, LJPL, GZL and MGC collectedthe data; GZL and MGC performed data analysis and interpretation, drafted themanuscript, and reviewed and approved the final version for submission. Allauthors have read and approved the final manuscript. The authors declare noconflicts of interest and confirm that all data underlying the findings arefully available without any restrictions.