This is the first study assessing malaria transmission heterogeneity in the Navrongo HDSS using a comprehensive entomological dataset and rigorous geostatistical and temporal models, which take into account data characteristics. These data indicate the presence of seasonal, spatial and year-to-year variation within a small geographical area (1,675 km2) in northern Ghana. The findings confirm previous studies reporting heterogeneity in malaria transmission in small areas. In particular, spatio-temporal variation has been observed in coastal Kenya
, in Kilombero valley in Tanzania
, in some selected Ugandan villages
 and in a low transmission zone in Sudan
Transmission in the Kassena-Nankana district is high (EIR > 100 ib/p/y) especially during the wet season. An entomological survey conducted in the same district between June 2001 and May 2002 recorded EIR of 630 ib/p/y in the irrigated zone within the southern part of the district
 which is lower than the one observed in the first year of the MTIMBA project. The drop of EIR after the first year may be explained by variations in laboratory testing. The ELISA tests for the first year were carried out in a different laboratory from those in the remaining two years, making it possible that inter-laboratory differences contribute to inter-annual variation. The year effect included in the model is, therefore, aliased with any laboratory differences. Consequently, there will be more confidence in EIR comparisons between locations than those that depend on inter-annual differences.
This study confirmed the presence of An. funestus and An. gambiae malaria vector species in the region
[16–18], with both acting as major vectors. NDVI, distance to water bodies and temperature were associated with mosquito density for both species. Compounds located close to water bodies were more likely to have high mosquito densities. The Kassena-Nankana district in northern Ghana has many irrigation dams that were constructed to increase food production in the area. There are also many small dugout reservoirs in the area which supply water to various communities especially in the dry season
[31, 32]. These water bodies can be favourable breeding grounds and responsible for mosquito abundance in neighbouring compounds. The data showed that a reduction in day temperature favoured higher number of mosquito bites in the area. The NHDSS where data were collected experiences high temperatures in some months (18°C to 45°C). Temperatures close to 40°C reduce mosquito survival, hence their density
. Although rainfall had a positive relationship with mosquito density, the association was not statistically important. However, rainfall is known to have a direct relationship with other factors, such as vegetation, that were found to positively influence mosquito abundance. A positive correlation between precipitation and mosquito density for both An. funestus and An. gambiae has already been observed in other places.
A seasonal pattern in mosquito density was observed for both species. High mosquito densities were observed in the rainy season for all the three years and low densities during the dry season. However, SR was higher in the dry than the rainy season during the second and third year. In addition, An. gambiae SR in the dry period were higher than that of An. funestus for the entire survey period. There was no evidence of variations in SR between species in the rainy season. More infected mosquitoes during dry seasons have already been observed in other areas
. This implies that most surviving adult mosquitoes in dry seasons are likely to be infectious.
The shortest distance at which the spatial correlation was below 5% was lower for SR than mosquito densities, suggesting that SR depends largely on local conditions rather than environmental factors. On the other hand, mosquito densities had strong spatial correlation and therefore they are more likely to be driven by environmental factors, especially vegetation which was the major predictor in the Navrongo area. Climate and environmental factors influence malaria transmission and its effects. In this district, malaria illnesses and mortality are observed thought the year with peaks in the wet season
[34, 35]. Blood transfusion, especially in young children, due to anaemia is more common in the rainy season
The EIR maps clearly depict spatial heterogeneity despite the relative small size of the HDSS. The high EIR estimate in the southern part, which is mainly covered by irrigation dams, has been reported previously
. Even during the dry season, transmission in the area remained high. In addition, the geographical pattern of EIR was similar across the three years of the project. The spatial and temporal variances of the mosquito density data accounted for about 33% each out of the total variation. However, SR data explained 67% and 33% of the total variation, suggesting that spatial heterogeneity was twice as high as the temporal one. Although space-time heterogeneity could explain total variation of the SR data, there was a remaining 34% unexplained variation for the densities. In principle, focussed malaria control conducted in the knowledge of these patterns of variation might be more effective than generalized intervention programmes, but no intervention programme is likely to be able to adapt to variations on this scale.
Bayesian geostatistical models are the state-of-art methodology to analyse space and time heterogeneity in malaria transmission and have been used to assess malaria risk using prevalence data
[37–41]. However, entomological data have large number of zeros, which cannot be estimated by standard geostatistical models. In particular, the Navrongo data had 44% and 72% of locations with zeros for density and SR, respectively. Entomological data were sparse in the other two MTIMBA sites (i e, Rufiji and Kisumu). This problem was addressed by developing geostatistical zero-inflated formulations of binomial models (GZIB) for analysing SR
. Zero-inflated analogues of negative binomial models
[22, 23] were also applied to take into account excess zeros in the density data. These models were able to improve EIR predictions obtained from standard geostatistical analogues.
The EIR estimates of this study will be used further to analyse the relationship between malaria transmission intensity and mortality as part of the ongoing work for the MTIMBA project.