Predicting distribution of malaria vector larval habitats in Ethiopia by integrating distributed hydrologic modeling with remotely sensed data
Location of potential larval habitats and probability of occurrenceGenerally, Anopheles arabiensis mosquito takes around 15 days to develop from egg to adult, but the duration can be as short as 10 days due to selection pressures from the stressed environment such as drought, temperature anomaly, or limited food resources48,49. In this regard, we considered areas with WI exceeding 10 and 15 days to be potential larval habitats under critical and normal conditions, respectively.To determine the probability of potential larval habitat occurrence, we computed the probability of ponding occurring longer than 10 and 15 days, P(WI > T), as shown in Eq. (2). P(WI > T) is defined as the ratio of D(WI(x,y,t) > T), the number of cumulated days for which the WI (i.e. persistence of ponding) of a grid cell (x,y) at time t that exceeded T days, to Dperiod, the number of days within a defined period of simulation.$$Pleft( {WI > T} right) = frac{{D(WIleft( {x,y,t} right) > T)}}{{D_{period} }},,T in left{ {10,15} right}$$
(2)
Figure 5 shows the results for the spatial distribution of P(WI > T) over the three periods of simulation, namely the entire year of 2018, the dry season (i.e. January to April and November to December) and the rainy season (i.e. May to October). It can be observed that ponding was persistent throughout the year around the stream edges and the vicinity. P(WI > 10) and P(WI > 15) were consistently close to 1, reflecting a high potential of these areas as larval habitats.Figure 5Spatial distribution for the probability of potential larval habitat occurrence. (a–d) represent the probability of WI exceeding 10 days and 15 days for the baseline scenario and the irrigation scenario for the entire year. Similarly, (e–h) represent the probability of WI exceeding 10 days and 15 days during the dry season, and (i–l) represent the probability of WI exceeding 10 days and 15 days during the rainy season. Areas where the simulated surface water flowrate exceeded 0.01 m3/s for 90% of the time in the simulated year were masked out for all the sub-figures since Anopheles larvae have a lower chance of surviving in fast-moving water61.Full size imageFor the baseline scenario shown in Fig. 5a,b, the P(WI > T) for the areas outside of the streams was predominantly determined by soil type. The areas characterized by Usterts (see Supplementary Fig S2) with the lowest hydraulic conductivity in the model domain were the next most at risk, with a P(WI > T) of about 0.4–0.5. In the remaining areas, P(WI > T) was generally 0.2 or less. Comparing Fig. 5a,b, the differences were minimal except for the steep areas at the watershed upstream boundary where P(WI > 15) was predominantly zero. The surface water ponding was unable to last more than 15 days due to the terrain gradient.Figure 5c,d show the results for the irrigation scenario. Compared to the baseline scenario, the year-round persistent ponding around the streams and the vicinity was wider in coverage and more noticeable. Irrigation also increased P(WI > 10) in Fig. 5c and P(WI > 15) in Fig. 5d from 0.4–0.5 to about 0.7 and 0.6 respectively for Farm #1, Farm #2, and a significant portion of Farm #3 and Farm #4. The P(WI > T) for the remaining area within the farms remained relatively unchanged at 0.2 and this could be attributed to the Ustoll soil type which drains easily. The increase in the probability of potential larval habitat occurrence from the baseline was more pronounced for P(WI > 10) than P(WI > 15) since the interval of irrigation was set at 10 days, after which the farm drained without replenishment until the next irrigation cycle.For the dry season, it can be observed in Fig. 5e,f that the stream edges were the only areas with high potential of larval habitat occurrence. In Fig. 5g,h, P(WI > T) increased visibly in the farms after irrigation, with a distinct similarity between Farms #1/#3 and Farms #2/#4 that points to the irrigation schedule. While irrigation was alternated evenly between the two groups, Farms#1 and #3 showed a higher P(WI > T) than Farms #2 and #4, possibly due to the timing of the irrigation relative to the rainfall. Irrigation could either coincide with rainfall or act as a supplement when there was no rainfall to augment soil moisture. Noticeably, there was an area to the northeast straddling both Farm #3 and Farm #4 where P(WI > 10) was around 0.1 but P(WI > 15) was almost zero, indicating that irrigation only allowed for larval habitats under critical conditions in that area during the dry season.For the rainy season, it can be observed in the baseline scenario (Fig. 5i,j) that the areas characterized by Ustert exhibited a high potential of larval habitat occurrence, apart from the stream edges. Particularly, there was an area to the north where P(WI > T) was lower than the other parts which could be due to the relatively steeper terrain. In the irrigation scenario (Fig. 5k,l), there was no visible difference in P(WI > T) as compared to the baseline scenario, apart from a minor increase around the western part of Farm #4.As a summary, we present the results in boxplots as shown in Fig. 6 to illustrate the effect of irrigation in different seasons for the areas inside and outside farms. The relevant statistics can be found in Table 1. The P(WI > T) had a highly asymmetrical distribution because it was very low in most of the model domain but could be very high in the remaining areas due to the streams. For the following comparison, we will use the median as it was more representative of the distribution.Figure 6Box plots for the probability of potential larval habitat occurrence for the whole year, dry, and rainy season. Probability of WI exceeding (a) 10 days and 15 days (b) for the area inside farms and the area outside farms. The line within each box is the sample median and the top and bottom of each box are the 25th and 75th percentiles. The whiskers were drawn from the two ends of the box and demarcate the observations which were within 1.5 times the interquartile range from the top and bottom of the box.Full size imageTable 1 Summary statistics of the probability of potential larval habitat occurence for the whole year, dry season, and rainy season. Mean, 25th percentile (P25), median and 75th percentile (P75) of the probability of WI exceeding 10 days and 15 days for the (a) areas inside farms and (b) areas outside farms. The p value was derived from the Wilcoxon Rank-Sum test under the null hypothesis that irrigation did not increase the median probability of exceedance from the baseline scenario.Full size tableIn the baseline scenario, there was a higher potential for larval habitats to form inside the farms, with a median P(WI > 10) of 0.427 and a median P(WI > 15) of 0.400, than outside the farms, with a median P(WI > 10) of 0.06 and a median P(WI > 15) of 0.019. This is expected because the farms are located in an area with relatively flat terrain and a higher concentration of streams. The difference in the median P(WI > T) inside and outside the farms was bigger in the rainy season compared to the dry season, as the higher rainfall received intensified ponding.Irrigation increased the median P(WI > T) inside the farms drastically in the dry season, with the median P(WI > 10) increasing from 0 to 0.442 and the median P(WI > 15) increasing from 0 to 0.282. Although irrigation was only applied over the dry season, there was also a statistically significant increase in the median P(WI > T) during the rainy season (p 10) increased from 0.848 to 0.864 while the median P(WI > 15) increased from 0.794 to 0.810. This was due to irrigation contributing to the antecedent soil moisture before the onset of the rainy season, which shortened the time for the soil to become saturated and ponding to occur. On the other hand, there was no strong evidence outside the farms of an increase in the median P(WI > T) due to irrigation (p > 0.01). This applied to both rainy and dry seasons.Stability of larval habitatsIn the previous section, we showed that irrigation did not have a significant impact on areas outside the farms. Here, we evaluated the stability of the potential larval habitats specifically for the areas inside farms based on the distribution of the maximum duration of ponding for each grid cell within the year as shown in the histogram (Fig. 7a). The total number of cells corresponding to each bin interval of 15 days was expressed as a fraction of the total number of cells in the area inside farms.Figure 7The fraction of area inside the irrigated farms for each potential larval habitat types under the baseline and irrigation scenarios. (a) Shows the histogram of the maximum duration of ponding within the year for the grid cells in each type of habitats expressed as a fraction of the total area of the farms. The bin size was 15 days. Temporary, semi-permanent, and permanent larval habitats were typically characterized by ponding duration of 15–90 days, 90–180 days, and 180 days and above, respectively. The baseline scenario is represented in blue and the irrigation scenario is represented in orange. The darker orange bin is a result of the two overlapping. (b) Shows the comparison of the fractional area occupied by non-habitats (less than 15 days) as well as potential temporary, semi-permanent, and permanent larval habitats inside the farms. Each grid cell within the farm was categorized based on its maximum ponding duration.Full size imageFrom the baseline scenario, 13.2% of the area was not favorable for larval habitats because the maximum duration of ponding in those areas was less than 15 days. The most common maximum ponding duration was between 150 and 165 days, which accounted for more than 20% of the area. This was followed by 15–30 days and 360 days and above which made up 17.6% and 13.8% of the area respectively. With irrigation, there was a general increase in the maximum ponding durations. The most common maximum ponding duration was 360 days and above, accounting for 18% of the area. Noticeably, the area with maximum ponding duration between 210–225 days increased fourfold to 10%. The remaining increase was for 285 days and above. Counter-intuitively, the area that was not conducive as larval habitats (i.e. maximum ponding duration less than 15 days) also increased slightly by 0.6%. This was because irrigation raised the overland flowrate in these areas, mostly near streams, and made them unfavorable for breeding.In Fig. 7b, we grouped the maximum ponding durations into stability periods corresponding to temporary (2 weeks to 3 months), semi-permanent (3–6 months), and permanent (6 months and above) habitats based on field observations from a study at the site35. Temporary habitats such as puddles retain water for a short period while permanent habitats such as stream edges and swamps hold water much longer and are more stable. For the baseline scenario, semi-permanent habitats were the most common, occupying 33.1% of the area, while permanent and temporary habitats also accounted for 29.6% and 24.1% of the area respectively. After irrigation, there was a significant shift from semi-permanent habitats, which reduced to 22.9% of the area, to permanent habitats which increased to 41% of the area. There was also a slight reduction in the extent of temporary habitats to 22.4% of the area.Temporal pattern of potential larval habitatsTo shed light on the temporal patterns, we evaluated F(WI > T), the fractional coverage of potential larval habitats inside farm, for each day throughout the year. We only focused on the area inside farms since irrigation does not have a significant impact on the area outside farms. As shown in Eq. (3), F(WI > T) is defined as the ratio of C(WI > T), the number of cells for which the WI (i.e. persistence of ponding) exceeded T days, to Cfarm, the number of cells within the farm area. T is set as 10 days and 15 days, corresponding to critical and normal conditions respectively.$$Fleft( {WI > T} right) = frac{{Cleft( { WIleft( {x,y,t} right) ge T} right)}}{{C_{farm} }},,T in left{ {10,15} right}$$
(3)
In Fig. 8a, F(WI > 10) increased steeply on January 10 as WI started increasing from 0 at the beginning of the year. For the baseline scenario, the fractional coverage decreased minimally from 0.18 throughout the dry season despite the sporadic spike in precipitation. At the onset of the rainy season, the peak rainfall event of the year from May 5th to May 11th caused a sharp increase in F(WI > 10) from 0.15 to 0.61 and thereafter, the relentless rainfall maintained the fractional coverage at about 0.6. Throughout the rainy season, there were four recurring peaks at a frequency of about 2 months. Post-rainy season, F(WI > 10) dropped gradually to below 0.2 after the last peak at the end of October.Figure 8Daily variations in the extent of the potential larval habitats for the year. Time series of the fractional coverage of areas with Wetness Index (WI) exceeding (a) 10 days and (b) 15 days.Full size imageFor the irrigation scenario, F(WI > 10) increased during the dry season from January to March with visible cyclical variations between 0.2 and 0.4 due to the rotation of irrigation among the four farms. Subsequently, the spike in rainfall at the end of March combined with the higher antecedent soil moisture from irrigation brought forward the step increase in the fractional coverage to April from May in the baseline scenario. As irrigation stopped at the end of April, F(WI > 10) gradually dropped back to the same level as the baseline scenario at the end of June. In the dry season from November to December, the fractional coverage started to deviate from the baseline scenario again with cyclical fluctuations, gradually decreasing towards the end of the year.In Fig. 8b, F(WI > 15) remained largely the same for the dry season but the peaks were moderated in the rainy season, compared to F(WI > 10). There was one less peak at the end of May in the early rainy season because the watershed did not accumulate enough rainfall for the persistence of the ponded areas to exceed 15 days. Specifically, for the irrigation scenario, the increase in fractional coverage during the dry season was moderated and less sensitive to the spikes in rainfall. Similarly, irrigation resulted in the early onset of the steep increase in F(WI > 15) in April following the spike in rainfall at the end of March. Also, it took two months after the end of irrigation in April for the fractional coverage to return to the same level as the baseline.From F(WI > 10) and F(WI > 15), we calculated the corresponding monthly mean, MF(WI > 10), and MF(WI > 15) as well as the 95th confidence interval as shown in Fig. 9. In Fig. 9a, MF(WI > 10) in the baseline was the highest for the months between June and September, constituting a four-month window in which at least 50% of the area was conducive for larval habitat formation. Of the four months, the highest monthly mean fractional coverage was in July at 79.9%. Irrigation extended the window to include the months of April and May. The monthly mean fractional coverage increased 4.5 times to 64.3% in April and 1.4 times to 63.7% in May. The MF(WI > 10) for the rest of the months in the window (i.e. June to September) remained one of the highest but the increase due to irrigation was not statistically significant (p > 0.01). July remained as the month with the highest monthly mean fractional coverage at 80.0%. In Fig. 9b, MF(WI > 15) was generally slightly lower than MF(WI > 10) for both the baseline and irrigation scenarios but the general trends were the same.Figure 9Monthly variation in the extent of the potential larval habitats for the year. Monthly mean fractional coverage of areas with a probability of WI exceeding 10 days (a) and 15 days (b). The 95% confidence interval is indicated at the top of each bar chart. The asterisks (*) next to the month on the x-axis indicate that irrigation increased the fractional coverage of the potential larval habitats for the month from the baseline scenario based on a 2-sample t-test (p More
