Where environment and malaria intersect: Exploring the spatio-temporal footprints of Plasmodium falciparum in Tanzania

TL;DR

This study explores how environmental factors influence malaria caused by Plasmodium falciparum in Tanzania over 20 years.

Contribution

The study identifies key environmental predictors of P. falciparum malaria in Tanzania using spatio-temporal analysis.

Findings

Kigoma, Kagera, Lindi, and Mtwara regions were consistent hotspots for P. falciparum from 2000 to 2020.

Precipitation and aridity were the most important environmental factors influencing malaria incidence.

Insecticide-treated nets and wet days also significantly impacted P. falciparum prevalence.

Abstract

Malaria remains a public health crisis in Tanzania, with persistent morbidities and mortalities. Malaria etiology is multifactorial, with environmental factors playing a vital role in mosquito development and malaria transmission. In Tanzania and most of Sub-Saharan Africa, the Plasmodium falciparum parasite remains the most prevalent and virulent malaria parasite. Using data from the Tanzania Demographic and Health Surveys and spatio-temporal analysis, we explore the environmental determinants of P. falciparum across different regions in Tanzania over the last 2 decades. The hotspots analysis showed that the Kigoma and Kagera regions in the north-west of Tanzania as well as the Lindi and Mtwara regions in southern Tanzania were consistently hotspots of P. falciparum malaria from 2000 to 2020. Our findings also reveal and reinforce the role of environmental factors in mediating malaria epidemiology in Tanzania. Factors such as the use of insecticide-treated nets, population, evapotranspiration and aridity were often adversely associated with P. falciparum incidence. In contrast, vegetative landcover, temperature, precipitation, and the number of wet days were directly associated with P. falciparum in Tanzania. However, the relationship between these environmental factors and malaria prevalence varied temporally and spatially. Our findings further showed that, the two most important environmental factors that mediate P falciparum incidence in Tanzania over the last two decades were precipitation and aridity. Other vital predictors included the use of insecticide nets and the number of wet days. The findings provide policy pointers for targeted malaria interventions in Tanzania in the context of environmental change.

Figures

Map of Tanzania.

Spatial distribution of malaria incidence in Tanzania from 2000–2020.

Getis-Ord Gi* hotspots analysis of malaria incidence in Tanzania from 2000–2020.

Ordinary least square (OLS) regression of the association between covariate and malaria incidence from 2000 to 2020.

Standardized beta coefficients of the environmental covariates of P. falciparum from 2000 to 2020.

a. MGWR of Aridity and malaria incidence, b. MGWR of precipitation and malaria incidence, c. MGWR of population and malaria incidence, d. MGWR of EVI and malaria incidence, e. MGWR of Rainfall and malaria incidence, f. MGWR of PET and malaria incidence, g. MGWR of ITN usage and malaria incidence, h. MGWR of number of wet days and malaria incidence, i. MGWR of temperature and malaria incidence.

MGWR of the intercepts of malaria incidence in Tanzania.

Full text

Introduction

One of the critical public health threats, particularly in Sub-Saharan Africa (SSA), is the ongoing challenges of malaria control, despite the efforts towards reducing and eliminating the prevalence [1,2]. Indeed, SSA accounts for 95% of all global cases of malaria and 96% of all malaria deaths, with children under the age of 5 accounting for 80% of these deaths [1]. Despite, the total funding for malaria control and elimination reaching 3 billion dollars, the prevalence of malaria remains far from elimination in SSA [3]. Tanzania is among the seven countries in the World Health Organization (WHO) African Region with the highest malaria burden [4], with over 93% of the population still at risk of malaria [5]. While there has been a decline in malaria transmission recently, malaria continues to be a leading cause of morbidity and mortality in Tanzania [6]. To highlight, the WHO reports that Tanzania is one of four countries that accounts for just over half of all malaria deaths worldwide [7]. Vulnerable populations, such as pregnant women and children, are more likely to be disproportionately impacted by malaria. Malaria continues to remain responsible for up to one-fifth of deaths among pregnant women and more than one-third of deaths among children under the age of 5 [8].

There are a complex number of factors that contribute to malaria transmission. At the individual level, misconceptions surrounding bed nets and indoor residual spraying, being of lower socioeconomic status, residing in homes built of mud, and not adhering to treatment regimens are all factors that can drive malaria prevalence [6]. Also, residing further away from healthcare facilities and visiting traditional healers can lead to the persistence of malaria [5,9]. Further, in Tanzania, there is a lack of an effective malaria surveillance system, which prevents public health experts from creating targeted interventions for at-risk locations [8,10]. Environmental conditions, such as higher cropland cover, and stagnant water are also associated with increased malaria transmission in the context of Tanzania [11,12]. Evidence also indicate that malaria incidence is connected with ecological and climate variability in Tanzania [12]. For example, rainfall is associated with seasonal peaks of malaria transmission [13,14].

To exacerbate the preceding concerns, the main malaria parasite in Tanzania is the Plasmodium Falciparum, one of the most severe malaria parasites [15]. For example, the Plasmodium parasite species is considered the deadliest globally and accounts for more than 90% of the world’s malaria mortality [16–18]. The other major malaria parasites, including Plasmodium malariae, Plasmodium vivax, and Plasmodium ovale have low prevalence in Tanzania [15].

While the role of various environmental and climatic factors in malaria transmission has been acknowledged in the literature [19–21], a scant amount of studies have thoroughly explored the importance of geographical and temporal variations in associated environmental factors and the role of this research in the creation of malaria control and prevention programs at both local and national levels. Importantly, even within Tanzanian national boundaries climatic and ecological processes vary significantly [22]. Therefore, there is a need to consider how local-specific climatic factors impact differences in malaria prevalence, as this information will be useful for public health policy. To this end, we contribute to the literature by exploring the geographical heterogeneity of environmental correlates of malaria prevalence in Tanzania in the last two decades. Findings from this study will help provide insights into malaria control and prevention by highlighting how environmental factors and variations can be incorporated into malaria control policy planning. Further, the findings will be useful not only in Tanzania, but in other countries in SSA that are striving to achieve Sustainable Development Goal 3, with a particular focus on Target 3.3, which states “End the epidemics of AIDS, tuberculosis, malaria and neglected tropical diseases and combat hepatitis, water-borne diseases and other communicable diseases [23].”

Environmental determinants of diseases

Patterns of infectious diseases have been associated with environmental changes across time and space [24]. For instance, there has been an increased focus on how climate change and urbanization can increase the prevalence of infectious diseases such as malaria, dengue fever, cholera and Lyme disease [25–27]. Hence, it is imperative to consider and utilize frameworks that help explain the underlying mechanisms of environmental factors impacting disease transmission.

One such example is the Environmental Determinants of Diseases (EnvID) framework. Made up of three interlocking components: environment, transmission, and disease [28], the foundation of this framework states that environmental processes impact the transmission cycles of infectious diseases. The environmental component has been disaggregated into distal environmental changes and proximal environmental characteristics. Distal environmental changes, deemed to be larger changes on a spatio-temporal scale, are those that impact disease transmission through multiple steps [28]. Examples include climate change, antibiotic usage, agricultural intensification, deforestation and urbanization [28]. These changes impact disease patterns through a series of casual linkages [28]. For example, climate change can impact precipitation levels, which may increase rainfall levels, creating stagnant water in certain areas, which increases the breeding and growing grounds for mosquitoes, which may carry malaria, yellow fever, dengue and West Nile Fever [29]. Proximal environmental characteristics are directly measurable components of the environment that directly impact the environment of the organisms and may directly affect the transmission cycle of an infectious disease [28]. Examples include temperature, precipitation, population density and humidity [28]. In some instances, it is not possible to distinguish between distal changes and proximal environmental characteristics, as it presents a continuum and is spatio-temporally dependent [28]. Thus, it is imperative to focus on proximal environmental factors as they are measurable, which can help in describing the relationship between environmental factors (such as rainfall and temperature) and infectious diseases (such as malaria). By establishing the relationship between these measurable factors and malaria prevalence in different regions of Tanzania, policy directions can become focused and specific, further targeting and benefiting those most at risk.

Utilizing the framework will assist in providing an overarching conceptualization to better understand the relationships between environmental factors and malaria prevalence in Tanzania. To reduce the burden of malaria in Tanzania, public health experts need to incorporate the risk factors of malaria into policy and interventions, focusing on those who are residing in the most at-risk areas and/or most vulnerable.

Materials and methods

Study area

The United Republic of Tanzania is a country located in East Africa (Fig 1). Geographically, it is located at 6.3690^◦^ S, and longitude of 34.8888^◦^ E, bordering Uganda, Kenya, the Indian Ocean, Mozambique, Malawi, Zambia, Rwanda, Burundi, and the Democratic Republic of Congo. Spatially, Tanzania covers a land area of about 945,087 km^2^. Composed of 30 administrative units, the population of Tanzania is about 62 million [30,31].

Map of Tanzania.

Tanzania is divided into four main climatic zones; the humid coastal plain; the semi-arid zone of the central plateau; the high-most lake regions; and the temperate highlands area [31]. In terms of temperature, the average temperature in Tanzania ranges between 27 degrees to 29 degrees along the coast and in the offshore islands, while temperatures range between 20–30 in the central, northern and western parts [32]. The average annual rainfall is 600–800 mm, with long rains occurring between March and May and shorter rains from October to December in the northern part of the country, with the rest of the country experiencing rain from December to May [22,33]. In the central part of Tanzania, the annual rainfall is approximately 550mm, while in the south-western highlands, the annual rainfall is about 3690 mm [32].

As a result of climate change, there will likely be changes and variability in temperature and precipitation. For example, in the 2050s the average annual temperature is expected to increase by 1–3 degrees [34]. Further, precipitation is projected to become more unpredictable, with changes in rainfall quantity and the onset of the rainy season(s) [35]. Indeed, research has shown that while the interior regions will see a decrease in precipitation by up to 20%, increasing the risk of drought, other areas of Tanzania will see an increase in rainfall, increasing the severity and frequency of floods [22,36].

Description of data

This paper is based on secondary data collected as part of the Demographic Health Survey (DHS) Program with support from the United States Agency for International Development (USAID). The data was acquired following a written request. We used the DHS Geospatial Covariates data and GPS locations cluster data (n = 628) from the spatial data repository. GPS locations of clusters are recorded during data collection in the field, and verified to ensure that they are within the right administrative boundaries in Tanzania [37]. However, to protect the confidentiality and privacy of the study participants (i.e., people showing clinical symptoms of P**lasmodium falciparum malaria), the DHS cluster GPS locations are geomasked by displacing the cluster by up to 2 km and up to 10 km for urban and rural clusters, respectively [37]. The geomasked points do not fall outside the administrative boundary of their associated clusters [38]. A detailed description of the DHS Geospatial Covariate and GPS data can be found in [37]. The DHS data adheres to the ethical standards of the ICF Institutional Review Board and country-specific (i.e., Tanzania) ethical guidelines. Participants’ confidentiality was ensured by displacing survey coordinates. Informed consent was read to each participant, and only participants who consented proceeded to answer the survey questions.

Measures

The outcome variable is the annual average clinical cases of P. falciparum malaria incidence in a DHS cluster location. Clinical cases of P. falciparum malaria can be described as malaria-attributable febrile episodes (i.e., a person’s body temperature beyond 37.5 C), which usually comes with nausea, fatigue, headaches, sweating and censored by a 30-day period. The occurrence of multiple sessions of these systems within the same 30-day period is classified as a single episode [37].

To explore the spatio-temporal association between the environment and P. falciparum malaria incidence, we used 9 environmental factors including aridity, temperature, rainfall, the number of wet days, population, Enhanced Vegetation Index (EVI), Potential Evapotranspiration (PET), precipitation and the use of Insecticide Treated Nets (ITN). Table 1 shows the detailed description of these variables.

Table 1: Description of the environmental covariates.

Analytical approach

The analysis included preprocessing the data, such as linking the geospatial covariate cluster data to the GPS cluster points and subsequently linking them to the Tanzania district boundaries. The coordinates of the cluster were projected from the WGS 1984 Geographic Coordinates System to the Arc 1960/ UTM zone 35S (EPSG:21035 with transformation: 1122) for analysis. The data analysis was implemented in R Studio version 2023.12.1 Build 402 and ArcGIS Pro version 3.2.0. The analysis included descriptive statistics, the use of choropleth maps, hotspot analysis (Getis-Ord Gi*), Ordinary Least Square (OLS) regression, and Multiscale Geographically Weighted Regression (MGWR).

Hotspots and cold spots analysis (Getis-Ord Gi*).

The Getis-Ord Gi* hotspots and cold spots analysis was used to map spatial clusters of high and low values of P. falciparum malaria incidence in Tanzania using a fixed distance band. A fixed distance band computes each feature considering the neighbouring features with use of a binary spatial weighting system [41]. Neighbouring features within the Euclidean distance are assigned a weight of 1, and features outside the Euclidean distance are assigned a weight of 0. In this paper, we utilized the Getis-Ord Gi* to compute output classes with z-scores, p-values and confidence intervals (i.e., 99%, 95% and 90% confidence intervals). Higher and positive z-scores depict statistically significant P. falciparum hotspots while negative and low z-scores indicate statistically significant cold spots of P. falciparum. The absolute value of the z-score depicts the intensity of clustering. Getis-Ord Gi* analysis identifies different intensities of clustering at various confidence levels. The Getis-Ord* analysis is computed as follows.

$[eqn]$ (1)

$[eqn]$ (2)

$[eqn]$ (3)

Where,

Gi* is the spatial autocorrelation statistic of an event i over n events.

Wj = spatial weight between i and j

n = is the total number of data points

xj = characterizes the magnitude of the variable x at events j over all n

S = standard deviation

Multiple ordinary least square (ols) regression.

The multiple OLS regression was used as a global model to assess the linear relationship between predictors and P. falciparum incidence in Tanzania. OLS is a global model that utilizes a linear equation in assessing the association between a linear outcome (e.g., malaria incidence) and a set of explanatory factors. OLS assumes a stationary relationship between the outcome variables and predictors, the result of which is the generation of a single coefficient for each variable [42]. The OLS regression is mathematically expressed as

$[eqn]$ (4)

Where;

y = the response variabley = the response variable (i.e., P. falciparum incidence)

xn = explanatory variables (aridity, ITN, EVI, PET, temperature, precipitation, rainfall, population, wet days)

β_0_ = Intercept

β_1_ = Parameter estimate of explanatory variable one

ɛ = Standard error

Multiscale Geographically Weighted Regression (MGWR).

Traditional global regression modeling such as OLS assumes that the association between variables are constant across a study area (e.g., Tanzania). However, spatial processes such as environmental factors may vary across geographic contexts, thus the use of global regression models will lead to misspecification because of the application of a constant value across all locations in the study area [43]. In GWR, spatial weights are assigned to the regression coefficient to generate different local coefficients across locations [44]. In GWR, a local regression equation is generated for each spatial unit, this allows spatial variation of the regression coefficients in the study location. The GWR can be described mathematically as follows. Given n observations, for the observation i ∊ {1,2,…,n} at locations i (ui, vi), the GWR model is given as

$[eqn]$ (5)

Where;

β_j_ (u_i_, v_i_)x_ik_ is the jth coefficient

Ɛ_I_ is the error term, and y_I_ is the outcome variable

u i, v_i_ are coordinates of geographic location i in space

Whereas GWR restrict the local associations within each regression model to vary at the constant spatial scale, MGWR permits the conditional associations between the outcome variable and the different explanatory variables to differ at different spatial scales [45,46]. The bandwidths indicating the data-borrowing range can change within the parameter surfaces [45]. In doing so, MGWR can provide vital insights into the scale at which different environmental processes operate, making it more flexible for assessing multiscale associations by relaxing the assumption that spatially varying phenomena in a model occur at the same spatial scale [45]. The MGWR is mathematically expressed as

$[eqn]$ (6)

Where bwj and βbwj depicts the bandwidth utilized in the calibration of the jth conditional association.

Results

The results are organized into four main sections. The first section provides basic description (e.g., mean, standard deviation, minimum and maximum values) of the outcome and explanatory variables. The second section uses choropleth maps to display the spatial distribution of malaria incidence from 2000 to 2020. The next section shows spatial and statistically significant clusters of malaria incidence hotspots and cold spots at different confidence intervals in 2000, 2005, 2010, 2015 and 2020. The last section highlights the global and local determinants of malaria incidence using Ordinary least square, and multiscale geographically weighted regressions.

Descriptive statistics

Table 2 presents the descriptive statistics of all variables used in this paper across all the years (i.e., 2000, 2005, 2010, 2015 and 2020). The average Malaria incidence decreased from an average of 0.83 and a maximum of 22.08 in 2000 to an average of 0.18 and a maximum of 4.93 in 2020 (see Fig 2). Also, the use of insecticide treated nets increased from an average of 0.03 (maximum = 0.08) in 2000 to 0.46 (maximum = 0.87) in 2020 (see Fig 2). In 2000, the mean aridity, temperature, EVI, precipitation, rainfall, PET, Wet days, and population were 19.79, 24.31, 0.30, 72.83, 898.90, 3.71, 10.05, and 414.97, respectively. In 2020, average aridity, temperature, EVI, precipitation, rainfall, PET, Wet days, and population were 29.64, 24.07, 0.35, 108.42, 1367.70, 3.67, 12.11, and 737.24, respectively.

Spatial distribution of malaria incidence in Tanzania from 2000–2020.

Table 2: Descriptive statistics of malaria incidence and environmental covariate from 2000 to 2020.

Spatial distribution of malaria incidence from 2000 to 2020 in Tanzania

The spatial distribution of malaria incidence from 2000 to 2020 are shown using choropleth maps (Fig 2). In 2000, the Lindi region in southern Tanzania had the highest malaria incidence and consistently through 2005, 2010, 2015 and 2020. The Katavi region in West-North Tanzania, the Mbeya and Singida in central Tanzania and Manyara in Eastern Tanzania had the next highest proportion of their populations with recorded clinical cases of P. falciparum malaria. In 2005, parts of the Katavi and Ruvuma regions had one of the highest recorded clinical cases of P. falciparum malaria aside from the Lindi region. By 2020, the number of areas with high malaria incidence per population had declined aside from a few areas in the Katavi region and Manyara regions in West-North Tanzania and Eastern Tanzania, respectively. The Liwale DC in the Lindi region remained the district with the highest malaria incidence per population.

Malaria incidence hotspots and cold spots in Tanzania from 2000 to 2020

The Getis-Ord Gi* Hotspots Analysis shows statistically significant clusters of hotspots (Fig 3) and cold spots of malaria incidence at different confidence levels (e.g., 99%, 95% and 90% confidence intervals). Districts in the Kigoma and Kagera regions in the north-west of Tanzania and in the Lindi and Mtwara regions in southern Tanzania consistently had significantly higher clusters (hotspots) of malaria incidence from 2000 to 2020 at 99% confidence interval. Also, districts in the Arusha, Kilimanjaro, and parts of the Manyara regions in the north-east of Tanzania had significantly low clusters (cold spots) of malaria incidence at 95% confidence from 2000 to 2020. Districts in the Mbeya region in the south-west of Tanzania also had significantly low clusters (colds spots) of malaria incidence in 2000, 2010 and 2015 at either 99% or 95% confidence intervals.

Getis-Ord Gi hotspots analysis of malaria incidence in Tanzania from 2000–2020.*

Environmental determinants of malaria incidence from 2000 to 2020

This section discusses the environmental determinations of malaria prevalence emphasizing first, the global level association between these environmental factors and malaria incidence using the OLS and standardized coefficients.

Global model results: OLS regression and predictive strength of environmental covariates.

The results of multiple OLS regression analysis between environmental factors and malaria incidence in Tanzania is shown in Fig 4. The results indicate that, an increase in the rate of ITN usage was significantly associated with a decrease in malaria incidence in 2020 and 2005, and inversely associated with malaria incidence in 2010. Also, an increase in aridity significantly decreased malaria incidence in 2015, 2005 and 2000. However, in 2020, aridity was positively associated with malaria incidence. EVI was significantly and positively associated with malaria incidence in 2015 and 2005. Similarly, temperature was significantly and positive related with malaria incidence in 2005 and 2000. Precipitation was also positively associated with malaria incidence in 2015, 2005 and 2000. Rainfall was significantly associated with malaria prevalence in 2020, 2015, 2010, 2005 and 2000, albeit its negligible coefficient. Similarly, as the number of wet days increased, malaria incidence increased, indicating a positive relationship.

Ordinary least square (OLS) regression of the association between covariate and malaria incidence from 2000 to 2020.

To understand the strength of each environmental factor in predicting P. falciparum, we used to standardize beta coefficients (Fig 5). Aridity (2020[β = 2.45], 2015[β = -1.28]) and precipitation (2020[β = -2.26], 2015[β=1.47]) were consistent the two strongest environmental predictors of P. falciparum in 2020 and 2015. In 2010, precipitation and the number of wet days were the two strongest environmental predictors of plasmodium falciparum. Also, precipitation and aridity emerged as the predictors of P. falciparum with the highest beta coefficients in 2005 and 2000.

Standardized beta coefficients of the environmental covariates of P. falciparum from 2000 to 2020.

Local model results: Multiscale geographically weighted regression (MGWR).

We explored the local relationships between the predictor variables and malaria incidence in Tanzania from 2000 to 2020. Results from the local MGWR is shown in Fig 6. Results from the MGWR showed no statistically significant local variation in the effect of aridity (Fig 6A) and precipitation (Fig 6B) on malaria prevalence from 2000 to 2020. Population (Fig 6C was significant and negatively associated with malaria incidence but showed no local variation in 2000 and 2005. EVI (Fig 6D) was significantly associated with malaria incidence with some local variation in 2005 and 2015 (showing a strong positive relationship in the central districts) and no local variation in 2000, 2010 and 2020.

a. MGWR of Aridity and malaria incidence, b. MGWR of precipitation and malaria incidence, c. MGWR of population and malaria incidence, d. MGWR of EVI and malaria incidence, e. MGWR of Rainfall and malaria incidence, f. MGWR of PET and malaria incidence, g. MGWR of ITN usage and malaria incidence, h. MGWR of number of wet days and malaria incidence, i. MGWR of temperature and malaria incidence.

Similarly, rainfall (Fig 6E) showed a statistically significant association with malaria in 2005 with no local variation and some local variation in 2010 (showing a strong inverse relationship in northern Tanzania and a moderate inverse relationship in central Tanzania). PET (Fig 6F) also showed a negative and statistically significant association with malaria incidence in 2010 and 2015, especially in northern Tanzania. ITN usage (Fig 6G), the number of wet days (Fig 6H), and temperature (Fig 6I) were significantly associated with malaria incidence with somewhat consistent local variation between 2000 and 2020. For example, the number of wet days had significant local variation in throughout the years. Temperature had significant local variations in the association with malaria incidence in 2000, 2015 and 2020.

The predicted values of malaria incidence without the influence of any environmental factor is shown in Fig 7. The results showed statistically significant and local variations in 2000, 2005, 2010, 2015 and 2020.

MGWR of the intercepts of malaria incidence in Tanzania.

Discussion

Malaria is a significant public health concern in sub-Saharan Africa (SSA), with Tanzania being one of the heavily burdened countries. Our study explores the spatial and temporal dynamics in the environmental determinants of P. falciparum malaria prevalence in Tanzania. Our study contributes to malaria epidemiology in Tanzania in three ways. First, we show the significant cluster of P. falciparum incidence in Tanzania. Second, we highlight the environmental predictors of P. falciparum incidence in Tanzania as well as identifying the most important environmental predictors in the last two decade. Third, we explore the local spatio-temporal variation in the association between environmental factors and P. falciparum incidence in Tanzania. This approach provides a comprehensive understanding of the spatial and temporal distribution of malaria and its environmental drivers to inform the development of targeted malaria control interventions and policies.

Our findings showed significant clusters of P. falciparum incidence in the regions of Kigoma and Kagera in the north-west of Tanzania over two decades, which is consistent with findings from [47,48]. For example, between 2007 and 2008, malaria prevalence was estimated to be 40% among children under five in the Kagera region [47]. Kagera has historically been a hotspot of malaria incidence, experiencing a severe malaria outbreak between 1997 and 1998 [47,49]. The population of Kagera is highly rural (about 94%, representing 5% of Tanzania’s population), partly explaining the high incidence of P. falciparum malaria incidence. In rural areas, the malaria avoidance behaviour that characterizes urban centres might not be so prevalent in rural areas, exposing populations to mosquito bites. Unlike rural housing infrastructure, urban housing structures may restrict mosquito access [50]. Furthermore, mosquito breeding is also more prevalent in areas with high agricultural land use, such as Kagera [8,51]. Similarly, our findings showed significant clusters of P. falciparum malaria incidence in southern Tanzania’s Lindi and Mtwara region, which concur with [3,52]. The clusters of P. falciparum malaria in these regions may be explained by the lack of indoor residual spraying [3] and the prevalence of biomodel rainfall, leading to high malaria transmission [53]. Moreso, the study revealed that the districts of Arusha and Kilimanjaro, and parts of the Manyara regions in the north-east of Tanzania had significantly low clusters (coldspots) of malaria incidence in the last two decades. This finding concurs with previous studies indicating that malaria prevalence was as low as 5% and 3% in northern and central parts of Tanzania, respectively compared to the southern and north-western Tanzania with malaria prevalence of about 33%-38%[3,54],

Consistent with the literature on environmental determinants of malaria in SSA [9,11, 12,16,55,56], our findings revealed that environmental factors such as temperature, aridity, landcover (i.e., vegetation), evapotranspiration, precipitation, rainfall, the use of ITN usage and the number of wet days were significantly associated with P. falciparum incidence in Tanzania. The use of ITN, population, PET and aridity were often adversely associated with P. falciparum incidence. The findings revealed that as ITN usage increased among the population, the incidence of P. falciparum reduced significantly in Tanzania. These findings corroborate the vital role of insecticide treated mosquito nets in SSA [16,57], with empirical evidence from more than 80 clinical trials all indicating that ITN usage prevents malaria transmission and mortalities [58]. The inverse relationship between population and malaria incidence can be explained by a higher human-to-mosquito ratio and the mosquito avoidance behaviour of urban centres with higher populations [16]. Also, higher evapotranspiration and aridity rates can reduce the level of ponds and puddles that serves as breading grounds for mosquito, thus reducing P. falciparum malaria incidence.

Among the environmental factors, EVI, temperature, precipitation, and the number of wet days were found to be directly related to P. falciparum prevalence in Tanzania. The relationship between vegetative landcover is consistent with findings from [12] that higher croplands and grasslands landcovers were associated with high malaria transmission in Tanzania. Also, increased temperatures were associated with increased P. falciparum prevalence, which is consistent with other studies that conclude that malaria increases with higher temperatures under favourable conditions [15,58]. Similarly, precipitation and the number of wet days were positively associated with P. falciparum prevalence in Tanzania. This association can be explained by the role of ponds and puddles as breeding grounds for anopheles mosquito larvae. Increased precipitation and the number of wet days may translate to available stagnant water which favours mosquito breeding. Also, our findings further reveal that the most important environmental factors that mediate malaria incidence in Tanzania in the last 2 decades were precipitation, aridity, and the number of wet days. Identifying the most vital environmental predictors of malaria incidence is a prerequisite for more targeted malaria prevention programs.

As highlighted in Nancy Krieger’s seminal work on “Epidemiology and the web of causation: Has anyone seen the spider?,” malaria epidemiology is mediated by biological, social and ecological factors, as such there was noticeable temporal (i.e., from 2000 to 2020) and spatial (i.e., across regions in Tanzania) variation in the environmental determinants of P. falciparum incidence. The association between ITN usage, wet days, temperature and P. falciparum showed significant spatial and temporal variability. For example, the association between temperature and P. falciparum incidence showed both temporal and spatial Variation. While temperature provides an optimal condition for mosquito larvae development, higher temperatures could also hinder malaria transmission [16]. According to Mordecai et al. [59], the optimal temperature for malaria transmission is about 25 ⁰ C, as such very higher temperatures may impede malaria larvae development. For example, higher temperatures could mean higher evapotranspiration thereby reducing stagnant water available for mosquito breeding. Similarly, while ponds and other stagnant water bodies are generally good breeding ground for mosquitoes, Mataba et al. [60] found that they might not be important factors in mediating mosquito borne diseases in certain areas of Tanzania such as the Manyara region.

While this study provides valuable insight into the spatio-temporal footprints and determinants of P. falciparum malaria in Tanzania, some limitations exist. We acknowledge that malaria epidemiology in SSA is complex and mediated by biological, socioeconomic, and ecological factors, all of which were not accounted for in the study due to data limitations. Moreso, the DHS data is a cluster-level dataset and does not measure individual biological, social, and economic characteristics that affect malaria transmission. Another limitation associated with the use of cluster data such as the DHS geospatial covariates data is Modifiable Areal Unit Problem (MAUP) [61]. The use of aggregated data may affect estimates at different scales [62]. To minimize MAUP, we used the smallest available administrative and geographic unit of analysis.

Conclusion

This study reinforces the multifactorial epidemiology of malaria in SSA by examining the spatial and temporal footprints as well as the environmental determinants of P. falciparum incidence in Tanzania. Our findings highlight that factors such as temperature, stagnant water availability and mosquito avoidance behaviours (ITN usage) significantly mediate malaria incidence. Importantly, the role of these factors in mediating malaria incidence varies temporarily and spatially. As such effective malaria prevention and mitigation in Tanzania ought to be based on an integrated and targeted framework. The primary interventions in Tanzania used by the National Malaria Control Programme (NMCP) mainly include indoor spraying, larvicide, insecticide nets, and prompt diagnosis tests [3]. There has also been mass social behaviour change communication to increase knowledge on malaria transmission, prevention and management.

However, in geographically and ecologically diverse country such as Tanzania, implementation of malaria prevention and mitigation intervention has not been adequately targeted [3], albeit micro variation in socio-cultural and environmental factors that mediate malaria prevalence. Our findings provide policy pointers for more targeted malaria interventions. First, there is a need for proactive malaria intervention that focus on mitigating favourable environmental conditions (e.g., stagnant water) for mosquito breeding and malaria transmission while scaling factors that hinder malaria transmission such as ITN usage. For environmental factors (e.g., populations) with no significant local variations in the effects on malaria incidence, the NMCP can employ national or general malaria intervention programs to mitigate malaria incidence. However, more locally targeted malaria intervention should be employed in regions with significant clusters of malaria incidence such as Kigoma and Kagera in the north-west of Tanzania as well as the Lindi and Mtwara regions in southern Tanzania. For localized malaria interventions, the NMCP can implement targeted interventions such as indoor residual spraying (IRS) and seasonal chemoprevention (SCP) in areas with high malaria prevalence. By implementing targeted policy initiatives, malaria control and prevention can avoid the ‘one-size-fits-all’ approach that may not meet the specific needs of high-risk populations and locations.

Appendix

Appendix 1, MGWR and GWR diagnostics

**: **