Background

Malaria control strategies have primarily focused on insecticide-treated nets (ITNs) and indoor residual spraying (IRS) to combat mosquito vector populations. However, these interventions are currently facing major challenges, including widespread insecticide resistance and behavioural adaptations of vector species [1, 2]. In response, there is need for complementary interventions, including larval source management (LSM), which is increasingly being considered by endemic countries, particularly in urban and peri-urban settings [1, 3]. In rural areas, LSM often faces various logistical challenges, particularly due to the complexity of aquatic habitats. Existing guidelines suggest that the aquatic habitats must be 'few, fixed and findable' [4], yet in most endemic regions this is rarely the case. The sheer numbers of breeding sites as well as their dynamic nature and inaccessibility illustrate the gap between LSM guidelines and their practical implementation in diverse settings.

Successful implementation of LSM programs requires a thorough understanding of the larval ecology of the target species, and especially the ability to locate their main aquatic habitats [5]. Whether malaria transmission is seasonal or perennial, identifying the main habitats that sustain vector populations throughout the dry seasons would be particularly important since such habitats could be targeted to maximize control when the vector populations are lowest [6].

Anopheles funestus sensu stricto (Anopheles funestus s.s.) is widely recognized as a major malaria vector in Eastern and Southern Africa [7, 8]. In south-eastern Tanzania [9, 10], as well as in some districts of northern Tanzania [11], this species is now responsible for > 85% of malaria transmission. This dominance is due to several attributes of this mosquito species, including its preference for both feeding on humans indoors and resting indoors [12, 13], its strong resistance to common pyrethroid insecticides [14] and its high daily survival rates [12]. Indeed, field evidence suggests that An. funestus can dominate malaria transmission even in areas where its densities are lower than those of other malaria vector species [9]. Unfortunately, in many settings, its basic biology and ecology are less well characterized compared to those of other vector species [8].

While studies focusing on An. funestus larval ecology are scarce, some of the studies carried out so far show that whereas its aquatic habitats occasionally overlap with those of other mosquito species, An. funestus possesses certain unique attributes that underlie its preferences [15, 16]. Early studies in the 1930s provided valuable insights, indicating that An. funestus was more likely to be found in permanent water bodies, such as river streams, ditches and ponds [17, 18], unlike An. gambiae complex mosquitoes, which generally prefer smaller and less permanent habitats [19]. A more recent study in south-eastern Tanzania found that An. funestus primarily oviposits in habitats along river tributaries, and in large ponds [17]. Distinctive features of these habitats, compared to those used by other malaria vectors, included clear waters, emergent vegetation, shading, water depths exceeding 0.5 m and permanent or semi-permanent availability [17]. Given the significance of An. funestus in the region, there is a need to extend these efforts by conducting detailed analyses of the importance of land cover characteristics.

The current study was therefore designed to explore how habitat characteristics, land cover types and human population densities affect An. funestus distribution, and then to use the findings to create habitat suitability maps for the vector species in south-eastern Tanzania.

Methods

Study site

The field survey was conducted in 18 villages located in south-eastern Tanzania, including 11 villages in Ulanga district (Chikuti, Chirombora, Ebuyu, Gombe, Ikungua, Iragua, Kichangani, Kidugalo, Lukande, Mwaya and Mzelezi) and seven villages in Malinyi district (Itete, Mtimbira, Sofi Mission, Sofi Majiji, Kalengakelo, Kiswago and Ipera Asilia) in south-eastern Tanzania (Fig. 1). The area has an altitude of 250–650 m a.s.l., yearly mean temperature ranges of 20–33 °C and annual rainfall range of 1200–1800 mm [20]. Generally, the dry season occurs between June and November, short rains occur in November and December, and long rains occur from February to May [20]. The area has diverse land use features, including small towns, villages, savannahs, crops, irrigation, grazing lands, forests and shrublands. There is a large flood plain with numerous rice farms, bordered by Udzungwa mountains to the north and Mahenge hills to the south (Fig. 1). Anopheles arabiensis and An. funestus are the main malaria vectors, with the latter species mediating most of the transmission [9, 21]. The main economic activities are livestock-kee**, fishing and crop farming [22, 23].

Fig. 1
figure 1

Study villages (filled red circles) for the dry season surveys of Anopheles funestus aquatic habitats in south-eastern Tanzania

Full size image

Sampling and characterization of aquatic habitats

The habitat survey was conducted during the dry months of September to December 2021. Community members aged ≥ 18 years were recruited and trained to identify potential aquatic habitats, including natural and human-made water bodies, and to record their physico-chemical attributes, regardless of whether mosquito larvae were present or not. To ensure a comprehensive coverage of each village, a team of five people walking at a distance of 2 m from each other systematically surveyed the area along pre-set transects, within demarcations set by the village authorities.

For each water body observed during the transect walks, the team recorded: (i) time and date of visit; (ii) GPS coordinates; (iii) habitat type (classified into river streams, stagnant ground pools, marshes, wells, dug pits, brick or concrete pits, ditches, rice fields, hoofprints); (iv) habitat size (surface area); (v) water clarity; (vi) water source (rainfall accumulation or ground water); (vii) water movement (stagnant, slow or fast moving); (viii) water permanence (permanent, semi-permanent, or temporary); (ix) water depth; (x) presence and type of algal growth (brown, blue, filamentous); (xi) presence of shading; (xii) types and quantity of vegetation; and (xiii) environmental characteristics surrounding the habitats within 200 m (such as cultivation, bush areas, cattle grazing and distance to nearest human habitations). Additionally, physico-chemical metrics, including pH, total dissolved solids (TDS) and electroconductivity (EC), were measured using a water-quality meter.

Larval surveys

All identified water bodies were examined for the presence of mosquito larvae using either the standard 350-ml dipper (for small habitats with shallow depths) or a large 10-l bucket (for larger and deeper habitats), as previously described [17]. The number of dips performed in each aquatic habitat was determined based on its size, following a predefined protocol. For habitats < 5 m2, a single dip was made; in habitats measuring 6 to 10 m2, two dips were made; and for those habitats ranging in size from 11 to 15 m2, three dips were made. This incremental approach continued for larger habitats, with a limit of 20 dips for any habitat > 120 m2. Collected larvae were identified to genus and species group level, whenever possible, using standard taxonomic keys [24, 25]. Within the Anopheles genus, late instars (III and IV) of the An. funestus group and the An. gambiae complex could be easily distinguished based on their morphology [25, 26]. Consequently, in this article, the term “aquatic habitats” refers to any surveyed water body, while “positive habitats” denotes those where An. funestus was confirmed via dip**.

Environmental covariates

A digital elevation model (DEM) with 10-m resolution [27] was used to extract data on elevation, slope, terrain and aspect for each aquatic habitat location. Land cover data were derived from the European Space Agency (ESA) Sentinel-2 satellite imagery acquired in June 2022. These data consisted of eight land cover classes and had a spatial resolution of 10 m, with and an overall accuracy of 75% [28]. The ESA imagery allowed analyses of both land cover and land use characteristics, such as urban areas and forestation, and helped identify small-scale landscape features and patterns crucial for understanding the local level relationships with malaria risk [29]. For each aquatic habitat, the proportion of each land cover type (water, trees, grasslands, flooded vegetation, shrubs, built-up areas, bare ground and crops) was extracted within a 300-m buffer. In addition, the distance from each aquatic habitat to the nearest feature of each land-cover class was measured. Consideration of both the buffer zone and distance to habitats allowed for a more nuanced analysis of how both the immediate landscape composition and the proximity to specific land cover types correlate with the presence of An. funestus larvae in the aquatic habitats.

Finally, human population densities data within the 300-m buffer were obtained from the Global Human Settlement Layer (GHSL) project, a spatial raster dataset composed of 100 × 100-m cells, with each representing the number of people in that area [30].

Statistical analysis

An initial descriptive analysis was conducted to assess the occurrence and distribution patterns of aquatic habitats occupied by An. funestus, as well as variations by type, specific location (village) and land cover categories (Fig. 2). A multivariate generalized linear model (GLM) with a binomial distribution was then used to examine the relationship between the presence of An. funestus larvae (absent = 0; present = 1) and a range of environmental and landscape variables (Table 1). Starting with a full model, including all the candidate variables, an automated backward stepwise selection was used to identify significant variables for inclusion in the final model, based on likelihood-ratio-tests (Table 1). To quantify the strength of association between the presence of An. funestus larvae and these variables, odds ratios (ORs) were calculated as part of the GLM. The ORs provide a measure of the likelihood of larvae presence associated with each variable.

Fig. 2
figure 2

Flowchart showing the analysis procedures, involving two datasets. On the left side, the boxes highlighted in green represent the field-collected data (dataset A), modelled using logistic regression to determine the significant variables and validated using the two-fold cross-validation technique. On the right side is a 200-m grid covering the entire study area (dataset B), which was used to perform the prediction of the habitat’s suitability based on the retained significant variables

Full size image
Table 1 Candidate covariates evaluated for predicting the presence of aquatic habitats of An. funestus mosquitoes
Full size table

A two-fold cross-validation process was used to validate performance of the model. The dataset was divided into a training subset (80%) and a test subset (20%). The model was trained on the training dataset and validated on the test set using Tjur's R2 calculations and area under the curve (AUC) receiver operating characteristics (ROC), with upper limits of 1.0 for a perfectly fitting model (Fig. 2).

The final model was used to generate spatial predictions of the likelihood of encountering An. funestus larvae in aquatic habitats found in different locations. To generate these maps, we created a 200-m resolution grid of regular points covering the study area of Malinyi and Ulanga districts (total area = 22,777 km2). Covariates retained in the final logistic model, such as proportions of land cover types, were extracted and applied to each grid to predict habitat suitability across the unsampled areas. To model how variations in specific habitat characteristics might influence the suitability for An. funestus, multiple scenarios were tested, with varied attribute values. For example, scenarios were created where water turbidity or habitat permanence were varied, reflecting different potential conditions.

All statistical analyses, including variable extraction, model fitting and predictions, were performed using the R statistical program version 4.2.1, with the packages rms, MASS, lme4 and glmm [31]. Preliminary data handling and visualization were performed using the software QGIS (Quantum Geographic Information System [32]).

Interactive maps for predictions and web application

To facilitate the exploration of different suitability scenarios, we also developed an interactive map using the Leaflet and Shiny packages in R [57]. Second, the accuracy of detecting mosquito larvae is influenced by sampling methods, including the number of samples, technical expertise and spatial coverage. Our study may have been limited by the pre-specified nature of these parameters. Finally, integrating field environmental data with remote sensing land cover data presents multiple challenges. For example, the resolution of land cover data used here might have been insufficient to capture fine details such as isolated residences and small water bodies, thereby impacting habitat suitability map**. To address these limitations, future studies may consider adopting higher-resolution remote sensing data sources, such as unmanned aerial vehicle (UAV) imagery [58,59,60], to capture finer details of habitats within a smaller geographical area and detect potential water sources for An. funestus [60].

Conclusions

This study comprehensively identified the main land use and environmental factors that influence larval habitat use by An. funestus during the dry season in southern Tanzania, where the species is the dominant vector of malaria transmission. We found that river streams and ground pools were the primary larval habitats during the dry season and that water bodies in forested areas, grasslands and shrublands are most likely to be positive for An. funestus. In contrast, larvae were least likely to be found in aquatic habitats which are in built-up and semi-urban areas. These insights are crucial for the strategic implementation of LSM strategies, particularly during the dry season when habitats are typically “few, fixed and findable.” The habitat suitability model developed here can be instrumental in pinpointing geographic areas where An. funestus larvae are most likely to be found, thereby facilitating targeted LSM deployment. Such targeted strategies, including larviciding and habitat modification, can be more effectively applied in high-risk zones identified through our model, thereby enhancing the efficacy of malaria control measures during the dry season. Building on these insights will further refine our understanding of mosquito dynamics, paving the way for enhanced strategies in malaria control and elimination.