Climatic conditions in Northern State were typically dry and warm to hot, depending on the season. During the period February 2004 to May 2007 the mean monthly temperature for the two field areas combined was 28°C and mean monthly minimum and maximum temperatures were 19.9 and 36.4°C, respectively. The hottest months were May-September, during which maximum daily temperatures consistently exceeded 40°C (Figure 3). The coolest month was January (mean minimum monthly temperature 10.7°C). Average daily diurnal temperature range varied between 14.9°C in August and 18.6°C in March and April. Mean monthly relative humidity was lowest in the period April-June (<18%) and highest in December and January (>36%). No rainfall was observed during the whole study period.
Over a period of 26 months, 52 field surveys were carried out in the Dongola and Merowe field areas. A total of 3,349 aquatic habitats were sampled, of which 321 (9.6%) contained An. arabiensis larvae (Table 1). The proportion of sites containing larvae differed significantly between the two field areas (Dongola = 12%, Merowe = 8.1%; p < 0.001). In Dongola the proportion of habitats positive for larvae was higher in static blocks than in random blocks (17.4 vs. 8.8%; p < 0.001). No significant difference between the two types of block was observed in Merowe (8.8 vs. 7.6%; p = 0.32). Late stage instars (L3+L4) were found in 77% of larvae-positive sites. Pupae were observed in 30% of larvae-positive sites.
Table 2 lists the types of aquatic habitat encountered in both field areas in descending order of frequency across both field areas. In Dongola the most common water sources encountered were canals (23% of the total sample), followed by containers (including zirs – earthenware pots – 10.8%), taps (10.7%) and riverbank (9.3%). In Merowe, containers (26%) and canals (23.5%) were by far the most common types of water sources sampled. The frequency with which larvae were found varied markedly by habitat type. In Dongola, aquatic habitats associated with 'grassy knolls' (soil terraces close to the main river channel), khors (seasonal tributary channels), leaking underground pipes, riverbanks, brickworks, seepage from canals and leaking surface pipes were most often positive for larvae. All wells and residue flood water pools sampled were positive for larvae, but together constituted a very small sample (Table 2). Larvae were absent from containers, cisterns and leaking water tanks and were found only rarely in canals, flooded fields, domestic drains, taps or culverts. A broadly similar picture emerges in Merowe, where habitats associated with grassy knolls, riverbanks, leaking underground pipes, brickworks, khors and leaking canals were most likely to contain larvae. Unlike Dongola, all wells in the Merowe field area were unproductive. Note that Table 2 represents a combined dataset including both random and static blocks. Separate analyses for static and random block showed similar patterns.
During field surveys, sites were also classified according to surrounding situation or land cover (see Table 3). The largest number of aquatic habitats was found in areas of settlement (1325, or 40% of all habitats visited), but fewer than 5% of these sites contained larvae. Similarly, of more than one thousand aquatic habitats surveyed within palm groves, only 6% contained larvae. Positive sites were most likely to be found in areas constituting channel edge (n = 370; 35% of sites positive) and within brickworks (n = 89; 19% of sites positive).
Logistic regression modelling for larvae presence/absence
Water body characteristics
Bivariate logistic regression models combining presence of An. arabiensis and various water body characteristics are presented in Table 4. In both field areas the presence of larvae was more likely in the following situations: in water bodies larger than 1 m2; in moderately saline water; in water of pH > 6.5; and in relatively un-shaded sites. The presences of algae or vegetation were significant risk factors for the presence of larvae in both field areas. In Merowe, larvae were more likely to be found in non-turbid water, but this distinction was not observed in Dongola. In Merowe, larvae were less common in deep water (depth > 50 cm), but again this effect was not evident in Dongola. Water temperature was not a significant determinant of risk of larval presence in either field area. For both field areas, minimum adequate models developed using multivariate logistic regression retained all significant variables (p < 0.001) in Table 4. In other words, none of the covariates significant in the bivariate analysis became non-significant when controlling for other covariates.
Two measures of land cover in the vicinity of aquatic habitats were generated in this study (Table 5). The first is a field-based assessment of situation or surrounding land cover type; the second corresponds to the RS-derived aggregated land cover class allocated to each 100 × 100 m survey block. The two classification systems, although not identical, are broadly comparable. For directly-observed land cover, and using 'bare fields' as a reference class, the presence of land cover types associated with the edge of the Nile channel (including rock pools, sand and mud banks, khors and grassy knolls) constituted the single strongest risk factor for the presence of larvae (for the two field areas combined, OR = 4.6; 95% CI = 3.0–7.0; p < 0.0001). Presence of settlement was associated with relatively low risk of larvae in both field areas. Risk of larvae in cultivated areas or brickworks did not differ significantly from the reference class in either field area. In Dongola, palm groves were associated with a lower risk of larvae than the reference class, but no similar significant effect was observed in Merowe.
Within the RS-derived land cover classification areas of settlement were arbitrarily sub-divided into localities within or outside a 200 m buffer of the Nile channel ('riverside settlement' and 'inland settlement' respectively). Using inland settlement as a reference class, areas close to the channel edge were associated with a greatly elevated risk of larvae being present (for the two field areas combined, OR = 8.2; 95% CI = 5.4–12.5; p < 0.0001). In Dongola, areas of riverside settlement were significantly more likely to contain larvae than areas of settlement 'inland' – but no significant difference in risk between the two classes was apparent in Merowe. In Merowe, land primarily given over to cultivation ('field-dominated mosaic') was more likely to contain larvae than the reference class, although the significance of this difference was borderline, and no similar effect was observed in Dongola.
To evaluate the overall effect of distance from river on presence of larvae (i.e. regardless of local land cover), an additional logistic regression model was developed on the basis of distances between each aquatic habitat and the Nile channel, calculated in a GIS (Table 5). Using habitats within 200 m of the channel as a reference class, in Dongola there is a clear and monotonic decline in risk with increasing distance from the Nile (and particularly for distances > 400 m). In Merowe the relationship between risk and distance to channel appears to be more complex -there being no significant increase in risk in areas greater than 800 m from the river.
Spatial distributions of larvae
The spatial distributions of surveyed sites in the two field areas are shown in Figure 2, which also shows the land use categories used to stratify the random survey blocks in each site. To assess whether distinct spatial clusters in the distribution of breeding sites exist, a spatial scan statistic was determined for randomly-selected sites separately for Dongola and Merowe. In Dongola, two clusters consisting of 16 breeding sites (expected = 2.3; relative risk = 8.7; p < 0.001) and 18 breeding sites (expected = 3; relative risk = 7.7; p < 0.001) were identified (Figure 2a). In Merowe (Figure 2b), one major cluster containing 12 breeding sites (expected = 1.9; relative risk = 7.1; p < 0.001) was identified at the north of the field area. Four very small clusters, each containing between four and seven breeding sites were also identified.
Seasonal distributions of larvae
Bar charts in Figure 3 show seasonal patterns in the proportion of aquatic habitats positive for An. arabiensis larvae, together with temporal variations in mean daily temperature and mean daily river gauge (river height) over the two field areas. Distinct seasonal patterns are evident, particularly in areas situated within 200 m of the main channel (Figure 3a) where decreased river height appears to promote breeding. Seasonal variations in areas away from the river are less marked, and percentages of larvae-positive sites are generally much lower in these areas, compared with sites within 200 m of the main channel. (Note however, that the total number of positive habitats found in inland and at riverside sites was roughly similar due to the much larger number of sites sampled inland). During very high river levels (August to November), the proportion of larvae-positive sites was higher in areas away from the river.
Seasonal patterns in the proportion of habitats containing larvae differed between years. In 2005, for example, the level of the Nile fell rapidly from mid-September and by December the river gauge, averaged across the two sites, was 10.8 m and the proportion of aquatic habitats in riverside areas containing An. arabiensis larvae was above 40% (Figure 3a). In 2006, river levels appear to have dropped off relatively slowly in the period October-December; the average river gauge in December was 1.3 m higher than it had been twelve months previously and the proportion of habitats positive for larvae in areas near the Nile was only about 10%.