Study site
The study was conducted in Smithfield, 15 km north of Cairns, Queensland, Australia (16.8221°S, 145.6972°E) (Fig. 1). The study site is situated proximal to an extensive swamp that provides larval habitats for a range of mosquito species near the Young Animal Protection Society (YAPS) Animal Refuge and the Smithfield Recycling Transfer Station. The specific location borders the recycling station and swamp forest with a tree canopy dominated by Melaleuca, Archontophoenix palms and Ceriops mangroves [9]. The closest house was ~500 m away. The site has a tropical climate with hot, humid summers and cooler, drier winters. Average annual rainfall is 1992 mm and temperatures range between 23–31 °C in the summer and 18–26 °C in the winter [10]. Although malaria was eliminated from Australia in 1981 [11], the former dominant vector, Anopheles farauti, is a common mosquito in northern Australia.
Barrier screen design and mosquito sampling
Barrier screens were constructed of 6 m straight lengths of 1.8 m high, high-density polyethylene (HDPE) UV stabilised shade cloth (Coolaroo® Gale Pacific Ltd, Melbourne, Australia) attached to poles and erected parallel to the swamp forest. On the side of the barrier screen opposite to the swamp, 150 g of dry ice was placed within a 2 litre cooler jug with four small holes to release CO2. This was then placed 1 m behind each barrier screen to simulate a blood meal source. Although not usual, the dry ice was used here to maximise collections on the barrier screens and to facilitate a more powerful direct comparison of the different construction attributes.
The impact of four basic barrier screen parameters on numbers of mosquitoes collected were evaluated individually as follows: shade cloth weight; shade cloth colour; construction design; and frequency of inspection. After the optimal shade cloth weight was determined by comparing different shade cloth weights of identical colour, that weight was then used to determine optimal colour. The optimal weight and colour was then used to evaluate optimal designs and inspection frequencies.
Cloth weight is defined by grams per square meter (g/m2) with green cloth of 135 g/m2, 160 g/m2 and 214 g/m2 tested, corresponding to 50%, 70% and 90% shading, respectively. The impact of colour was then evaluated using white, green and black cloths of optimal weight (determined as described previously). Barrier screen construction design was varied to determine if eaves of 25 cm depth could improve collection efficacy. It was hypothesized that mosquitoes would remain on the screens for longer periods by baffles or eaves; therefore, three barrier screen designs were evaluated using the optimal weight and colour cloth as previously determined. Screens without eaves (a straight 1.8 × 6 m flat shade cloth), screens with perimeter eaves (e.g. vertical eaves along both sides and a horizontal eave along the top) and a screen with complete eaves (e.g. vertical screens on both sides and three horizontal eaves at heights of 60 cm, 120 cm and 180 cm from the ground; Fig. 2). The frequency of inspections on mosquito numbers collected was evaluated by inspecting sets of identical screens (optimised for weight and colour as described above) at intervals of 30 min, 60 min and 90 min during 3-hour sampling periods.
A balanced Latin square design (3 × 3) was used to compare each of the experimental parameters. Triplicate barrier screens were erected in a row (separated by 2 m gaps). Each variable tested was rotated through all three potential spatial positions over three consecutive nights to eliminate any location associated bias (a full rotation of positions). For each variable three full rotations were completed unless specified otherwise.
Replicate barrier screens were examined for mosquitoes by a single collector from 18:00 h until 21:00 h using a mouth aspirator to remove resting mosquitoes from the screens. The collector applied mosquito repellent (active constituent 92.8 g/L Picaridin Aerogard®, Sydney, Australia) and unless stated otherwise, mosquitoes were collected hourly with each searching event lasting approximately 10 min per screen with the swamp side searched first. Mosquitoes from the same screen and hour were stored in separate labelled polyethylene terephthalate (PET) holding cups. The resting height on the screen, low (0–60 cm from the ground), middle (60–120 cm above the ground) or high (120–180 cm above the ground) was also recorded for 3 nights. All mosquitoes were morphologically identified to genera and sex [12]. The study was conducted between March 2016 and February 2018.
Cloth weight was tested over three rotations where light, medium and heavy cloth was compared, an additional 2 rotations with only light and medium weighted cloth (to ascertain if there was a significant difference in numbers of mosquitoes collected between light and medium cloth) was carried out. The influence of cloth colour on mosquito numbers was tested during four rotations over 12 nights: with white, green and black barrier screens.
Screens without eaves were initially compared to screens with perimeter eaves during 2 nights (1 rotation). During the initial testing period, differences between the two designs were not found so screens with complete eaves were added for an additional 2 rotations (6 more nights of testing). The impact of the frequency of inspection on mosquito numbers collected was tested over 9 nights (3 rotations with inspections at 30 min, 60 min and 90 min).
Statistical analysis
The effect of barrier screen variables on resting female mosquito densities was analysed with a Generalized Linear Mixed Model (GLMM) with a negative binomial distribution and a random factor for the rotation of the Latin square (glmer.nb; package lme4) with a sequential post-hoc analysis to clarify any statistical differences between the experimental factors (glht; package multcomp). By incorporating the random factor for rotation into the GLMM, the model accounts for natural fluctuations in mosquito densities observed during sampling periods while increasing the power of the model. Separate analyses were conducted for each experimental parameter and for the An. farauti group (although An. farauti is the dominant species, An. hinesorum is found in the study area) as well as for mosquitoes in the genera Aedes and Culex. This analysis was conducted using R statistical software version 3.1.2.