Modelling larval habitat preferences for Ochlerotatus punctor and Aedes cinereus

(1)

Modelling larval habitat preferences for

Ochlerotatus punctor and Aedes cinereus

Oswin van der Scheer

June 2019

Abstract

Current policy in the nature reserve ”Deurnsche Peel & Mariapeel” concerning the facilitation of bog land causes an increased water level. This creates a nuisance situation mainly caused by two types of biting mosquitoes. This studied created habitat models for both species in order to suggest improvements to current policy. This study found that open candle rush fields, flooded forests, marsh vegetation and large bodies of open water harbour the lowest numbers of these mosquito species. Fur-thermore this study showed that edge effects may cause the number of mosquitoes to increase. To improve policy the aforementioned vegetation types should be facilitated in combination with buffer zones to minimize edge-effects.

(2)

Introduction

In the past, it was common for industrialized countries to try to drain wetlands in order to make them more suited for agricultural practices (Mitsch, 1994). The Netherlands, with its extensive swamps, marshes and wetlands is no excep-tion in this regard. However, due to biodiversity goals set by European law, and predictions by the IPCC (2007) concerning wetter winters and more dynamic summers with longer droughts and heavier rains (Sch¨afer et al., 2004), this ap-proach changed. Policy makers now aim to re-wet areas in order to store more water and sustain target-species.

This re-wetting of nature reserves, can also cause an increase in certain un-desired species. By re-wetting these reserves more temporary ponds, swamps, marches and water storage basins are created, which can serve as breeding sites for biting insects (Verdonschot & Besse-Lototskaya, 2014). Some of these species can cause serious nuisance situations for inhabitants of areas close to these re-serves, as well as recreational users of these reserves (Verdonschot & Dekker, 2019).

If biodiversity goals are to be met, and nature reserves are to be prepared for a changing climate, while limiting nuisance caused by re-wetting policy, a thorough understanding of the specific habitat characteristics of nuisance caus-ing species is vital.

In order to achieve this understanding, this study aims to formulate the habitat preferences for two nuisance causing, biting mosquito species. Since the larval habitat covers most of the important life cycle processes including larval development, pupating and emergence, it is the most important aspect to study (Soleimani-Ahmadi et al., 2013).

This study aims to create a model that helps to get a more complete un-derstanding of the habitat preferences of the species: Ochlerotatus punctor and Aedes cinereus. This aim can be translated into the following research ques-tions:

Which habitat characteristics can be used to predict the abundances of O. punc-tor and A. cinereus larvae?

How do these characteristics translate into policy deployed by nature manage-ment?

The study was conducted in nature reserves around the village of Griendtsveen. Previous reports on mosquito nuisance already showed that two species of biting mosquitoes, O. punctor and A. cinereus, are the main cause of nuisance in this area (Verdonschot & Dekker, 2019). Verdonschot & Dekker (2019) found that in this area three factors primarily influence mosquito abundances.

First, weather plays a vital role, mostly due to the effects that dry weather has on the water level. Second, the changing hydrological characteristics of the reserve. And finally, the structure of the area, including vegetation and

(4)

eleva-tion differences that create a patchwork of habitats.

Previous studies conducted in this area, also measuring habitat characteris-tics, were limited in time and survey usually crossed the study area once. This study tries to extend on this work by sampling the area with a higher frequency and by deploying regression models to find key habitat parameters. To achieve this, the area was studied over a longer time period, namely during 4,5 weeks, crossing the entire research area three consecutive times.

The acquired knowledge in this study, can then help to deploy policy that can combine re-wetting practices for water management purposes, the achievement of biodiversity goals and limitation of nuisance caused by biting insects.

(5)

Materials & methods

Study site

This study was conducted in the ”Deurnsche Peel & Mariapeel” Natura 2000 reserve in the Dutch provinces of Limburg and Noord-Brabant. The samples were collected between the 1st of April and the 2nd of May in three different sampling rounds.

The field survey in this study aims to provide an area wide coverage of mosquito abundances in a specified sub-area of the reserve around the vilaage of Griendtsveen (see figure 1).

Figure 1: Satellite photo of the area around the village of Griendtsveen. The study area is given by the blue line. Green dots indicate sampling locations. The study area was divided into six sub-areas, each being surveyed in one day. The researchers studied sub-areas 3, 4 and 6 in three rounds, and sub-areas 1 and 2 four times in order to be sure that the peak abundance of mosquito larvae was measured. Sub-area 5 was studied in the first round, but was less accessible in round 2 and 3. This caused sub-area 5 to be absent in round 2 data and much smaller in round 3 data.

Measured parameters

In order to attain area wide coverage, two researchers crossed the area, noting every time they encountered a change in habitat. This could be a change in vegetation, but also a change in wetness of the area or the amount of shade. If

(6)

the habitat-zone contained surface water, this water was sampled using a 350 mL Clarke mosquito dipper. This was done five times per habitat-zone in order to incorporate replicates in the experimental design. The water samples were searched directly for mosquito larvae or pupae, which were then preserved in 75% ethanol in labeled plastic 50 mL bottles.

Previous studies on mosquito larval habitats of different species by Jenkins & Knight (1952) already showed strong correlation with vegetation and habitat type. Therefore, most of the parameters measured in this study are of this nature, including: herb layer vegetation, wood layer vegetation and amount of shade. Furthermore, the characteristics of the bodies of water also affect larval abundances (Soleimani-Ahmadi et al., 2013). This is captured in the measurement of water depth, permanence and the percentage of the habitat that is wet.

Conductivity and pH were measured in order to account for the preference of the studied species O. punctor for more acidic waters (Verdonschot & Dekker, 2019), and possible preference for more mineral rich waters. In their study on shallow lakes, Das et al. (2006) showed that conductivity is a useful indicator for Total Dissolved Solids (TDS), and will therefore be used in this study as an indicator for richer waters.

Finally the distance of every measuring site to the edge of the nature reserve was measured in order to account for possible edge effects such as agricultural runoff.

In every new habitat zone the following parameters were noted by one of the researchers. This was always the same researcher in order to make the estimates more reliable.

Sampling round, Sub-site, Date, X-coordinate, Y-coordinate

The X- an Y-coordinates were measured using a Garmin eTrex Vista HCx GPS-tracker.

Number of dips containing larvae or pupae

The number of dips that contained larvae or pupae was noted. If the habitat-zone did not contain any surface water, -1 was noted to indicate a dry sample. Water depth

The water depth was noted in one of seven possible categories: dry (0), soggy (1), sole deep (5), ankle deep (10), knee deep (50), deep (100). The numbers indicate an estimate of the depth in centimeters.

Herb vegetation 1

The dominant herb vegetation for the habitat-zone was noted, being one of the following: None, grass, peat moss, bentgrass, candle rush, ferns, brushwood, heather, marsh vegetation (e.i. reed) or other.

Herb vegetation 2

If there was another type of herb vegetation present, alongside the dominant one, it was also noted in one of the above specified categories.

(7)

Woody vegetation

The dominant woody vegetation for the habitat-zone was noted, being one of the following: None, birch, oaks, willows, alder, birch and oaks, flooded forest or other.

Amount of shade

The amount of shade cast by the top vegetation was estimated and noted in one of three categories: open, half-open or closed.

Permanence

The permanence of the surface water was estimated and noted in the following categories: Dry, puddles, marsh or pool.

Percentage of area that is wet

The percentage of the habitat-zone that was covered by surface water was es-timated and noted in one of the following categories: less then 5%, 25%, 50%, 75% or more then 95%.

pH

If surface water was present in the habitat-zone, the pH was measured with one decimal precision using HACH HQ40D Multi Meter.

Water conductivity

If surface water was present in the habitat-zone, the conductivity was measured in micro Siemens per meter using a HACH HQ40D Multi Meter.

The pH and water conductivity parameters were only measured in the sec-ond half of round 3 and in round 4 due to limited resources.

Both the herb layer vegetation parameters and the woody layer vegetation pa-rameters were combined into a single vegetation parameter in order to make the data more suited for analysis. Nine different vegetation types were distin-guished. Characteristic pictures for each vegetation type are shown in appendix A figure 4-12.

Bentgrass and birch forest: This category contains all sample sites containing bentgrass or grass as primary or secondary bottom vegetation and birch as top vegetation. sample sites containing peat moss as bottom vegetation were not included in this category.

Peat moss and grass: This category includes all sample sites containing peat moss as primary or secondary bottom vegetation in combination with grass or bentgrass. Top vegetation could be either birch, willow or nothing.

Candle rush field: This category includes all sample sites containing Candle rush as primary or secondary bottom vegetation. Top vegetation could be either birch, willow or nothing. Sample sites containing peat moss were not included in this category.

Peat moss and candle rush field: This category includes all sample sites containing peat moss and candle rush.

Open water periphery: This category includes all sample sites containing comments indicating that the sample site is the edge of a pond, lake, canal, or an overgrown ditch.

(8)

Open water: This category includes all sample sites containing comments indicating that the sample site is a canal, ditch, pond or lake.

Flooded forest: This category includes all sample sites containing flooded forest or alder as top vegetation and sample sites that had comments or char-acteristics indicating a flooded willow forest.

Mixed forest: This category includes all sample sites containing birch and oaks as top vegetation.

Marsh: This category includes all sample sites containing marsh vegetation as primary or secondary bottom vegetation.

Data preparation

The raw data for this project is not open to the public due to its sensitive nature to involved parties. A request to view the raw data repository can be sent to prof.dr.ir. PFM Verdonschot at piet.verdonschot@wur.nl.

Nineteen data points that did not fit in any of the categories above were left out of the analysis since these were singular occurrences that did not aid in creating an idea of habitat preferences.

For every data point the distance to the edge of the reserve was calculated using ArcGis 10.7 and added as a parameter. This was done using the ’near’ package in ArcMap. Map data was retrieved from the University of Amsterdam Geoportal (n.d.).

The larvae and pupae were determined in the lab using the determination guide by Becker et al. (2010) for larvae, and the guide by van Haren & Verdon-schot (1995) for pupae.

If the larvae were still highly underdeveloped they were only determined up to genus level. The pupae were determined to the genus level. If the genus was Aedes they were further divided into O. punctor and A. cinereus if characteris-tics were clearly present. Otherwise they were also determined up to the genus level. Note here that in the used determination key for pupae, Ochlerotatus was still a sub genus under Aedes, but after the work of Reinert et al. (2004) this was revised and Ochlerotatus is now a genus.

Determination was done by the author and Dorine T.B.M. Dekker at Wa-geningen University & Research. Dekker also checked some of the author’s samples to ensure quality.

Data analysis

The data was analysed using R 3.5.3. For the analysis, 96 samples that did not contain any surface water were left out. Seventeen samples that had missing data on the parameters or that showed incoherent parameters were also left out.

A logistic regression model was formulated assuming a quasi-Poisson dis-tribution to account for over-dispersion of the data. The total amounts of O. punctor and A. cinereus larvae and pupae per sample, served as the response variables, whereas the above mentioned parameters served as the explanatory variables.

(9)

First, a full model with all parameters was formulated for each species. Along with the main effects, one interaction effect, between vegetation and amount of shade, was added to the model. The model was tested for auto-correlation among the explanatory variables using a Durbin-Watson test.

This model was then analysed using a Chi-squared test ANOVA to discover variables that showed a significant contribution to explaining the data. Vari-ables that were not significant were then left out creating a reduced model. The full model was compared with the reduced model using an F-test ANOVA.

If the reduced model was not significantly worse than the complete model this model was used to continue.

The new model (i.e. the reduced model) was then further analysed look-ing at p-values and explained deviance. The least contributlook-ing parameter was removed from the model and the new, even further reduced, model was again tested against the full model using an F-test ANOVA.

If the reduced model was still not worse then the full model, the new reduced model was used to continue. This was done until a model was found that was not significantly worse then the full model with as little parameters as possible. The time component of the model was further analysed by using a Tukey post-hoc test to make pairwise comparisons between the sampling rounds. Fur-thermore, the full model was computed using data from only one sampling round to see if this would yield different results due to possible dependencies in the data among the sampling rounds. This was done using data exclusively from sampling round 2, since mean abundances of both species were highest in this sampling round.

Finally a correlation between the distance to the edge of the reserve and the water conductivity in a sample site was calculated to see if bodies of water around the edge of the reserve were richer then those further away from the edge. This was done using a Pearson’s correlation test.

(10)

Results

O. punctor

The full model showed significant trends for all main effects, except for Perma-nence. The model showed no significant trend for the interaction effect.

After removing the insignificant parameters, the new reduced model did not explain the data significantly different (ANOVA, F = 0,6234, p = 0.883).

Removing the next least significant term, ’Amount of shade’, also did not cause the model to lower the explanation of the data (ANOVA, F = 1,4001, p = 0,115).

Removing the next least significant term, ’Water depth’, did however cause the model to lower the explanation of the data (ANOVA, F = 2,1614, p = 0,001).

This leaves the most suitable model as it is reported in table 1, with five main explanatory parameters.

Table 1: Parameters for O. punctor larval habitat model including residual and explained deviance and degrees of freedom per parameter

Further analysis of the categorical variables created the model as shown in table 2. This table shows the coefficients for the separate categories of the variables. Only the significant coefficients are shown, containing five negative interactions. For the complete results see Appendix B.

(11)

Table 2: Estimates for all significant coefficients in the O. punctor habitat model. Negative interactions are shown in red.

Comparing the model containing pH and conductivity parameters to a model without these predictors yielded no significant differences (ANOVA, F = 0,2142, p = 0,8077). This model was not used for further analysis since the data set is smaller then the original one.

A. cinereus

The full model showed significant trends for the ’Sampling round’, ’Vegetation’ and ’Water depth’ parameters.

Removing all other parameters resulted in a model that explained the data equally well as the full model (ANOVA, F = 0,5708, p = 0,955).

Removing the least significant parameter, ’Sampling round’, did not cause the model to lower explanation of the data (ANOVA, F = 0,987, p = 0,486).

Subsequently removing the ’Water depth’ parameter did lower the explana-tion of the data (ANOVA, F = 1,5855, p = 0,022). The same happened when removing the ’Vegetation’ parameter (ANOVA, F = 1,6164, p = 0,014).

This leaves the most suitable model as it is reported in table 3, with two main explanatory parameters.

Table 3: Parameters for A. cinereus larval habitat model including residual and explained deviance and degrees of freedom per parameter

Further analysis of the categorical variables created the model as shown in table 4. This table shows the coefficients for the separate categories of the

(12)

variables. Only the significant coefficients are shown, containing three negative interactions. For the complete results see Appendix C.

Table 4: Estimates for all significant coefficients in the A. cinereus habitat model. Negative interactions are shown in red.

Comparing the model containing pH and conductivity parameters to a model without these predictors yielded no significant differences (ANOVA, F = 1,0904, p = 0,3417).

Time series

Figure 2 shows violin plots of the abundances of both species grouped per sam-pling round. For O. punctor round 2 and 3 differ significantly (Tukey, adjusted p = 0,047). For A. cinereus no significant differences were found. The highest values are not shown since this would make the graph unreadable.

Figure 2: Violin graphs of the abundances of O. punctor and A. cinereus per sampling round. A wider violin indicates more data points at that y-value. Red dots indicate mean values and horizontal lines indicate 2nd quartile.

The Durbin-Watson test was insignificant for both species (O. punctor : DW = 1,2906, p = 0,820 ; A. cinereus: DW = 1,8607, p = 0,770) indicating no auto-correlation among the explanatory variables. Analysis using only one sampling round yielded the same significant parameters as with all sampling rounds, for both species (see appendix D).

(13)

Edge effects

A significant correlation was found between the distance to the edge of the reserve and the water conductivity of samples (Pearson’s r = -0,288, p = 0,002), indicating that richer waters are found closer to the edge of the reserve (figure 3 .

Figure 3: Correlation between water conductivity and Distance to the edge of the reserve. Fitted line is shown in blue and 95% confidence interval is shaded.

(14)

Discussion & Conclusion

O. punctor

This study suggests that the larval abundances can be best explained using a habitat model including the following parameters: Vegetation, Sampling round, Percentage of habitat that is wet, Water depth and distance of the habitat to the edge of the nature reserve.

In more depth, the coefficients suggest that, in order to minimize O. punctor abundances, a couple of habitat characteristics are preferable.

First, more open vegetation types like ’candle rush field’ tend to host lower abundances of O. punctor.

Next we see that permanently wet habitat types like a flooded forest, also tend to host lower O. punctor abundances. This is not very surprising as it has been proposed by Verdonschot & Dekker (2019) that Aedes and Ochlerotatus mosquitoes need long-term temporary waters to reproduce.

Furthermore, even though the relationship is small, the data suggests that O. punctor can be found more, the closer you move to the edge of the reserve. Combining this finding with the correlation between the distance to edge and water conductivity, suggests that the richer waters on the edge of the reserve are the cause of the higher numbers, by providing more food for the developing larvae. This suggests that edge effects have an influence on O. punctor , and that minimizing those effects, by for example creating buffer zones, might de-crease the abundance of O. punctor.

Finally, figure 2 suggests that O. punctor can be found mostly in early spring and less in later spring.

Combining all these findings, this model suggests that, in order to minimize nuisance caused by O. punctor, policy makers should focus on facilitating the following type of habitat: Open candle rush fields in combination with perma-nently wet flooded forests. In combination with this policy, buffer zones should be installed to minimize edge effects.

A. cinereus

This study suggest that A. cinereus larval abundances can be explained by a fairly simple model containing only two parameters: Water depth and vegeta-tion.

The most important parameter is vegetation type, suggesting that ’Open water periphery’, ’Marsh’ and ’Flooded forest’ type habitats host lower abun-dances compared to ’Bentgrass and birch forest’. This means that apart from facilitating these types of habitat, this model does not provide more policy strategies to minimize nuisance caused by A. cinereus.

The ’Open water periphery’, ’Flooded forest’ and ’Marsh’ vegetation types are all habitat types associated with relatively wet areas. This does suggest that re-wetting of the ”Deurnsche Peel & Mariapeel” nature reserve is a policy that does not favor A. cinereus, and for that matter O. punctor, and that con-tinuation of this policy might reduce nuisance caused by these species.

(15)

Comparison

The fact that the model for O. punctor is a better fit (47% of null deviance ex-plained in O. punctor model vs 17% in A. cinereus model), suggests something is missing in the A. cinereus model. Upcoming studies should therefore focus on this species and try to incorporate other parameters in an attempt to create a more complete habitat model.

Time component

When looking at the experimental design in this study, it could be argued that the measurements in each sampling round are not independent of each other, and that a different analysis would be more suitable.

This can be refuted however, by the fact that the researchers never crossed the reserve in exactly the same route, and did not sample the exact same bodies of water. The researchers aimed to make an area wide coverage of the study area and sampled all bodies of water they encountered. Sometimes a body of water was not present in a second sampling round, new bodies of water appeared, or the characteristics (i.e. size, depth, shade) of the sampled water differed so much that the samples can be viewed as independent. This makes linking of samples between sampling rounds, which is needed for a repeated-measures analysis, impossible.

To further strengthen the argument of independence, the Durbin-Watson test proved that no auto-correlation exists among the explanatory variables, something you would expect if the sampling rounds were not independent.

Finally the test with data from only one sampling round showed the same parameters to be significant in the full-model analysis. This suggests that the effect of the multiple sampling rounds on the outcome of the experiment is small, and that including them only strengthens the explanatory value of the model.

Water chemistry

The correlation between water conductivity and distance to the edge of the re-serve suggest that edge effects may be of influence. This is further supported by the fact that distance to the edge of the reserves is a significant explanatory variable in the final O. punctor model. To further investigate the magnitude of these edge effects on larval abundances in a follow-up study, water conductivity data should be collected throughout the whole study period and collecting TDS data should also be considered.

Even though pH did not prove to be a significant indicator for the studied species, collecting pH data throughout the whole study period might change this finding and should therefore also be considered in follow-up studies.

Verdonschot & Dekker (2019) already showed that weather has strong in-fluence on mosquito abundances and that results may vary from year to year. This study was conducted in a relatively dry year. To create an even better un-derstanding of the habitat preferences of the studied species, the study should be replicated in a wetter year.

(16)

Finally, this study’s intensive field mapping proved to be suited to define habitat models for both species of nuisance causing biting mosquitoes in the ”Deurnsche Peel & Mariapeel”. The results indicate that this type of study can be used in many different areas with different species to investigate habitat preferences.

(17)

References

Becker, N., Petri´c, D., Boase, C., Lane, J., Zgomba, M., Dahl, C., & Kaiser, A. (2010). Mosquitoes and their control (Vol. 2). Springer.

Change, I. P. O. C. (2007). Climate change 2007: the physical science basis: summary for policymakers. Geneva: IPCC .

Das, R., Samal, N. R., Roy, P. K., & Mitra, D. (2006). Role of electrical conductivity as an indicator of pollution in shallow lakes. Asian Journal of water, Environment and pollution, 3 (1), 143–146.

Jenkins, D. W., & Knight, K. L. (1952). Ecological survey of the mosquitoes of southern james bay. American Midland Naturalist , 456–468.

Mitsch, W. (1994). Wetlands of the old and new worlds: ecology and manage-ment. Global wetlands: old world and new , 3–56.

Reinert, J. F., Harbach, R. E., & Kitching, I. J. (2004). Phylogeny and classifi-cation of aedini (diptera: Culicidae), based on morphological characters of all life stages. Zoological Journal of the Linnean Society, 142 (3), 289–368. Sch¨afer, M. L., Lundstr¨om, J. O., Pfeffer, M., Lundkvist, E., & Landin, J.

(2004). Biological diversity versus risk for mosquito nuisance and disease transmission in constructed wetlands in southern sweden. Medical and Veterinary Entomology, 18 (3), 256–267.

Soleimani-Ahmadi, M., Vatandoost, H., Hanafi-Bojd, A.-A., Zare, M., Safari, R., Mojahedi, A., & Poorahmad-Garbandi, F. (2013). Environmental characteristics of anopheline mosquito larval habitats in a malaria endemic area in iran. Asian Pacific journal of tropical medicine, 6 (7), 510–515.

Uva geoportal. (n.d.). http://geodata.science.uva.nl:8080/geoportal/catalog/main/home.page. (Accessed: 2019-6-13)

van Haren, J. C., & Verdonschot, P. F. M. (1995). Proeftabel nederlandse culicidae (Tech. Rep.). Instituut voor Bos-en Natuuronderzoek.

Verdonschot, P. F., & Besse-Lototskaya, A. A. (2014). Flight distance of mosquitoes (culicidae): a metadata analysis to support the manage-ment of barrier zones around rewetted and newly constructed wetlands. Limnologica-Ecology and Management of Inland Waters, 45 , 69–79. Verdonschot, P. F., & Dekker, D. T. (2019). Stekende insecten griendtsveen

(18)

Appendix A: Vegetation types

Figure 4: Bentgrass and birch forest

Figure 5: Peat moss and grass

(19)

Figure 7: Peat mos and candle rush field

Figure 8: Open water periphery

(20)

Figure 10: Flooded forest

Figure 11: Mixed forest

(21)

Appendix B: O. punctor model

Estimate Std. Error t value Pr(>|t|) (Intercept) -7.0220 822.4990 -0.01 0.9932 Sampling round 2 0.3915 0.2672 1.47 0.1434 Sampling round 3 -1.5673 0.5060 -3.10 0.0021 Sampling round 4 -2.0892 0.9087 -2.30 0.0219 Vegetation 2 -0.5146 0.3583 -1.44 0.1516 Vegetation 3 -2.4144 1.0582 -2.28 0.0229 Vegetation 4 -1.5402 1.0523 -1.46 0.1438 Vegetation 5 -3.0060 1.7243 -1.74 0.0818 Vegetation 6 -0.3771 1.0004 -0.38 0.7063 Vegetation 7 -1.5055 0.7459 -2.02 0.0440 Vegetation 8 -0.5348 0.4613 -1.16 0.2468 Vegetation 9 -3.7640 2.1247 -1.77 0.0770 Percentage wet 1 -0.3280 0.3696 -0.89 0.3753 Percentage wet 2 0.3287 0.3336 0.99 0.3249 Percentage wet 3 -0.6925 0.4998 -1.39 0.1664 Percentage wet 4 -1.0784 0.6908 -1.56 0.1191 Distance to edge -0.0033 0.0009 -3.67 0.0003 Water depth 5 12.4781 822.4990 0.02 0.9879 Water depth 10 11.2109 822.4990 0.01 0.9891 Water depth 20 11.1918 822.4991 0.01 0.9891 Water depth 50 9.4246 822.5013 0.01 0.9909 Water depth 100 6.6871 822.5807 0.01 0.9935

(22)

Appendix C: A. cinereus model

Estimate Std. Error t value Pr(>|t|) (Intercept) -11.9478 722.7902 -0.02 0.9868 Vegetation 2 -0.0634 0.2133 -0.30 0.7665 Vegetation 3 -0.2800 0.3361 -0.83 0.4051 Vegetation 4 -0.2144 0.3415 -0.63 0.5304 Vegetation 5 -1.8581 0.7573 -2.45 0.0144 Vegetation 6 -0.9192 0.7997 -1.15 0.2509 Vegetation 7 -0.8383 0.3705 -2.26 0.0240 Vegetation 8 -0.7758 0.4301 -1.80 0.0718 Vegetation 9 -1.5911 0.5039 -3.16 0.0017 Water depth 5 14.7846 722.7902 0.02 0.9837 Water depth 10 14.0870 722.7902 0.02 0.9845 Water depth 20 13.6850 722.7902 0.02 0.9849 Water depth 50 13.7533 722.7903 0.02 0.9848 Water depth 100 12.1831 722.7929 0.02 0.9866

(23)

Appendix D: Model with single sampling round

O. punctor

Df Deviance Resid. Df Resid. Dev Pr(>Chi)

NULL 166 31264.38 Vegetation 8 9153.66 158 22110.72 0.0000 Distance to edge 1 2435.80 157 19674.92 0.0003 Amount of shade 2 3335.22 155 16339.70 0.0001 Water depth 4 3891.98 151 12447.71 0.0003 Permanence 3 290.03 148 12157.69 0.6640 Percentage wet 4 1844.46 144 10313.23 0.0397 A. cinereus

Df Deviance Resid. Df Resid. Dev Pr(>Chi)

NULL 166 3196.84 Vegetation 8 546.76 158 2650.08 0.0009 Distance to edge 1 0.11 157 2649.98 0.9428 Amount of shade 2 38.65 155 2611.33 0.3940 Water depth 4 283.26 151 2328.07 0.0085 Permanence 3 4.51 148 2323.57 0.9748 Percentage wet 4 130.29 144 2193.27 0.1792

Modelling larval habitat preferences for Ochlerotatus punctor and Aedes cinereus