The development of predictive tools for pre-emptive dengue vector control: a study of Aedes aegypti abundance and meteorological variables in North Queensland, Australia

Authors

  • Aishah H. Azil,

    1.  Sansom Institute for Health Research, University of South Australia, Adelaide, SA, Australia
    2.  Department of Parasitology and Medical Entomology, Universiti Kebangsaan Malaysia, Kuala Lumpur, Malaysia
    Search for more papers by this author
  • Sharron A. Long,

    1.  School of Public Health and Tropical Medicine, James Cook University, Cairns, Qld, Australia
    2.  Tropical Population Health Unit, Queensland Health, Cairns, Qld, Australia
    Search for more papers by this author
  • Scott A. Ritchie,

    1.  School of Public Health and Tropical Medicine, James Cook University, Cairns, Qld, Australia
    2.  Tropical Population Health Unit, Queensland Health, Cairns, Qld, Australia
    Search for more papers by this author
  • Craig R. Williams

    1.  Sansom Institute for Health Research, University of South Australia, Adelaide, SA, Australia
    Search for more papers by this author

Errata

This article is corrected by:

  1. Errata: Corrigendum Volume 20, Issue 8, 1125, Article first published online: 3 July 2015

Corresponding Author Craig Williams, Sansom Institute for Health Research, GPO Box 2471, Adelaide, SA 5001, Australia. Tel.: +61 883021906; Fax: +61 883022389; E-mail: Craig.Williams@unisa.edu.au

Summary

Objectives  To describe the meteorological influences on adult dengue vector abundance in Australia for the development of predictive models to trigger pre-emptive control operation.

Methods  Multiple linear regression analyses were performed using meteorological data and female Aedes aegypti collection data from BG-Sentinel Mosquito traps placed at 11 monitoring sites in Cairns, north Queensland.

Results  Considerable regression coefficients (R2 = 0.64 and 0.61) for longer- and shorter-term factor models respectively were derived. Longer-term factors significantly associated with abundance of adult vectors were mean minimum temperature (lagged 6 month) and mean daily temperature (lagged 4 month), explaining the predictable increase in abundance during the wet season. Factors explaining fluctuation in abundance in the shorter term were mean relative humidity over the previous 2 week and current daily average temperature. Rainfall variables were not found to be strong predictors of A. aegypti abundance in either longer- or shorter-term models.

Conclusions  The implications of these findings for the development of useful predictive models for vector abundance risks are discussed. Such models can be used to guide the application of pre-emptive dengue vector control, and thereby enhance disease management.

Abstract

Développement d’outils de prédiction pour la lutte préventive contre le vecteur de la dengue: une étude de l’abondance de Aedes aegypti et des variables météorologiques dans le nord du Queensland, en Australie

Objectifs:  Décrire les influences météorologiques sur l’abondance du vecteur adulte de la dengue en Australie afin de développer des modèles prédictifs permettant de déclencher des opérations de contrôle préventives.

Méthodes:  Des analyses de régression linéaire multiple ont été réalisées en utilisant des données météorologiques et des données sur des femelles de Ae. aegypti collectées à partir de pièges à moustiques BG-Sentinel placés sur 11 sites de surveillance à Cairns, au nord du Queensland.

Résultats:  Des coefficients de régression considérable (R2 = 0,64 et 0,61) pour des modèles de facteurs à plus long et plus court terme, respectivement, ont été tirées. Les facteurs de plus long terme significativement associés à l’abondance des vecteurs adultes étaient la température minimale moyenne (décalée de 6 mois) et la température moyenne quotidienne (décalée de 4 mois), expliquant l’augmentation prévisible de l’abondance pendant la saison des pluies. Les facteurs expliquant la fluctuation de l’abondance dans le court terme étaient l’humidité relative moyenne au cours des 2 précédentes semaines et la température quotidienne moyenne actuelle. Les variables des précipitations n’ont pas été trouvées comme de bons prédicteurs de l’abondance de Ae. aegypti, ni dans les modèles de plus long ni dans ceux de plus court terme.

Conclusions:  Les implications de ces résultats pour le développement de modèles prédictifs utiles pour les risques d’abondance du vecteur sont discutées. Ces modèles peuvent être utilisés pour guider l’application de lutte préventive contre le vecteur de la dengue et donc améliorer la prise en charge de la maladie.

Abstract

Desarrollo de herramientas predictivas para el control vectorial preferente del dengue: estudio de la abundancia de Aedes aegypti y variables metereológicas en North Queensland, Australia

Objetivos:  Describir la influencia de la metereología sobre la abundancia del vector adulto del dengue en Australia, para el desarrollo de modelos predictivos que desencadenen operaciones de control preferentes.

Métodos:  Se realizaron análisis de regresión linear múltiple utilizando datos metereológicos y de hembras de Aedes aegypti recolectadas en trampas de mosquitos BG-Sentinel puestas en 11 lugares de monitorización en Cairns, North Queensland.

Resultados:  Se derivaron coeficientes de regresión considerables (R2 = 0.64 y 0.61) para modelos de factores a largo y corto plazo. Los factores que a largo plazo estan asociados de forma significativa con la abundancia de vectores adultos son la temperatura mínima (en un lapso de 6 meses) y la temperatura media diaria (lapso de 4 meses), lo cual explica el aumento predecible en abundancia durante el período de lluvias. Los factores que explican la fluctuación en abundancia a corto plazo son la humedad media relativa en las 2 semanas anteriores y la temperatura media del día. Las variables de lluvia no son buenos vaticinadores de abundancia de Ae. aegypti, ni a corto ni a largo plazo.

Conclusiones:  Se discuten las implicaciones de estos hallazgos para el desarrollo de modelos predictivos útiles para determinar el riesgo de abundancia de vectores. Estos modelos pueden utilizarse para guiar la aplicación de control vectorial preferente para el dengue, y por lo tanto mejorar el manejo de esta enfermedad.

Introduction

Aedes aegypti is currently the only dengue vector on mainland Australia and is responsible for the almost annual transmission in northern Queensland since the re-emergence of dengue in Australia in the 1980s and 1990s (Kay et al. 1984; Ritchie et al. 2002). The region has been subject to significant epidemic dengue activity in recent times, with 909 locally acquired cases of DENV3 notified during the 2008–2009 wet season (Queensland Health 2009).

Dengue management in the region primarily consists of vector control responses to notified cases (Queensland Health 2005). However, some entomological surveillance with accompanying vector control response is currently carried out as a pre-emptive measure against transmission in some areas as is performed in other regions (Ooi et al. 2006; Eiras & Resende 2009). A key index measured in Queensland is the weekly mean adult female collection in BG-Sentinel traps; a surveillance programme conducted in the city of Cairns, which is subject to regular, seasonal dengue transmission secondary to extra-regional case importation.

In general, the link between entomological indices and dengue transmission is poorly understood (Scott & Morrison 2003), although some studies have sought to rectify this (Focks et al. 2000; Wu et al. 2007). In Australia, an association between collections in sticky ovitraps and dengue transmission was described (Ritchie et al. 2004), but beyond this there have been no formal analyses in this country.

Developing an understanding of dengue risk based on meteorological and entomological indices that can be applied sustainably and measured prospectively is a priority for dengue research (Morrison et al. 2008). If such risk factors can be inculcated into risk analysis tools, this will provide health authorities with quantitative measures that can be used to instigate pre-emptive vector control operations, an approach rarely used in dengue control (Morrison et al. 2008).

To determine how best to measure and predict entomological risk, we are engaged in projects that examine the utility of various entomological indices. Further, the ability to predict indices ahead of time would be useful for health authorities planning vector control responses. One way to achieve this is by clarifying associations between entomological indices and meteorological factors.

The importance of such an approach is apparent in numerous studies. Favier et al. (2006) for example, investigated the relationships between weather variables, larval and pupal indices and found that relative humidity and temperature were the only significant weather factors that correlated with these indices. Conversely, in Malaysia Wan-Norafikah et al. (2009) found that these two weather variables were not significant factors, rather rainfall was significantly associated with larval density. Positive associations (albeit not formally analysed) between weekly rainfall and adult A. aegypti abundance have been demonstrated in Sao Paulo, Brazil (Dibo et al. 2008; Fávaro et al. 2008), and in Buenos Aires, Argentina (Vezzani et al. 2004). Lag effects, whereby past rainfall events influence vector abundance in subsequent weeks or months, were demonstrated. Positive relationships between rainfall, temperature and humidity and dengue vector abundance are understandable. Rainfall will fill breeding habitats (e.g. anthropogenic containers) for mosquitoes; increases in temperature will accelerate mosquito larval growth; and increasing humidity will enhance adult mosquito survivorship, facilitating greater feeding on humans, dispersal and egg laying. However, each of these three variables may impact A. aegypti populations in slightly different ways, depending on local ecology. Thus, this diversity of relationships between entomological and meteorological indices highlights the need to investigate such relationships at local ecological scales.

The aims of the work described here were to characterize the links between meteorological variables and weekly BG-Sentinel trap collections. Specifically, we sought to identify the major meteorological influences that operate seasonally (i.e. long term) and to characterize the more proximal (i.e. short-term) influences on A. aegypti collection with a view to developing predictive tools for this disease risk factor.

Methods

Study site

Cairns Queensland (16°55′ S, 145°46′ E) has a resident human population of ca. 130 000 (CDATA online, Australian Bureau of Statistics 2009). The climate is tropical with distinctive hotter wet- (November–May, mean maximum temperature 30.2 °C, mean monthly rainfall 262.4 mm) and cooler dry seasons (June–October, mean maximum temperature 27.1 °C, mean monthly rainfall 35.1 mm). Dengue fever is epidemic in Cairns, usually following viral importation by travellers returning from dengue-endemic countries. In the past, outbreaks have occurred during the wet season, when vector abundance is relatively higher than in the dry season (Hanna et al. 2003).

Entomological data

Adult A. aegypti were collected using BG-sentinel™ mosquito traps (Biogents AG, Regensburg, Germany). This trap is an effective tool to attract and collect adult mosquitoes in tropical Australia (Williams et al. 2006, 2007). It out-performed several other collection methods in residential areas in Cairns (Williams et al. 2006). Furthermore, BG-sentinel™ mosquito traps can be used in rapid and routine assessment of dengue vector populations (Williams et al. 2007).

Weekly BG-sentinel trap collections were conducted at 11 monitoring sites (i.e. houses) from early September 2006 until mid-April 2009 (= 137 collections) in the suburbs of Parramatta Park, Bungalow, Manunda and Cairns North; all within 4 km of the city centre (Figure 1). Both female and male mosquitoes were collected from the traps, but only the numbers of females were included in the analysis. Mean female A. aegypti per day (henceforth meanfemday) was calculated by dividing the number collected at each monitoring site with the number of days the traps were set; mostly 7 days.

Figure 1.

 BG-Sentinel mosquito trap sites (denoted by a star) operated from September 2006–April 2009 in Cairns, Queensland, Australia.

Meteorological data

Meteorological data in five categories were obtained from Australian Bureau of Meteorology (http://www.bom.gov.au) for the nearest weather station (‘Cairns Aero’), approximately 6 km from central Cairns. These were as follows: (i) mean daytime temperature (°C) (TEMP), (ii) maximum daily temperature (MAXTEMP), (iii) minimum daily temperature (MINTEMP), (iv) total daily rainfall (RAIN) and (v) mean daily relative humidity (RH). Daily observations in these categories were used to calculate variables for inclusion in regression analyses. For each of the five categories, mean values were calculated for the week of mosquito trapping, along with a range of time-lagged periods, specifically 1, 2 and 3 weeks, and 1, 2, 3, 4, 5 and 6 months prior to each trapping week (i.e. meanfemday observations).

Multiple linear regression analysis

Stata/IC 11.0 for Windows (StataCorp LP, TX, USA) was used to perform multiple linear regression analysis. Linear relationships between meanfemday and all meteorological variables were examined in graph matrices. Any non-linear relationships were dealt with by transformation. Specifically, meanfemday and rainfall variables were transformed using log10 (x + 1) to meet the assumption of linear regression analysis. We ran the multiple regression analysis in two tiers: first by analysing all the meteorological variables and secondly by analysing only the shorter-term meteorological variables – lag 0 week, lag 1 week, lag 2 week, lag 3 week and lag 1 month.

To select the most significant variables in each meteorological group (TEMP, MAXTEMP, MINTEMP, RAIN and RH), two types of stepwise multiple regression analysis were performed: (i) inclusion criteria at P value ≤0.05 and (ii) exclusion criteria of P value ≥0.2. Subsequently, the significant variables from these two types of analysis were combined and included in second-stage regression analysis without applying any inclusion/exclusion criteria. Multicollinearity (as measured by variance inflation factor, VIF) was investigated at this stage and solved by discarding one or more variables that were highly correlated (VIF > 4) for each variable. Variables with the highest R2 (i.e. regression coefficients) in individual regression analysis were chosen amongst collinear variables. Then, all chosen variables from each meteorological group were subjected to further analysis. These variables were combined and analysed using similar methods as described earlier. Finally, meteorological variables that could best explain the variance in meanfemday were included in the regression models. Any variable with > 0.05 was excluded from the models.

Regression diagnostics were performed to investigate heteroscedasticity, normal distribution of residuals and presence of outliers. Remedial measures (dropping of significant outliers determined by Cook’s test) were taken to improve the validity of the regression models.

To validate Model 2, we used actual meanfemday data recorded from 23 April to 26 November 2009 (= 33) and performed a correlation between model-predicted and actual values. However, because correlation alone might not be adequate to assess the agreement between the predicted and actual meanfemday (Altman & Bland 1983; Bland & Altman 1986), we plotted the differences between each pair of these two values (Y axis) versus the average of each pair (X axis); a method introduced by Altman and Bland (1983).

Results

Aedes aegypti adult female abundance varies seasonally, with collections peaking in January–February and steadily declining to reach their nadir in August (Figure 2). This seasonal trend is reflected generally in weekly mean of average daytime temperature and humidity (Figure 2). Despite this broad seasonal trend, there is some weekly variation readily apparent. Furthermore, the population peaks vary in their timing over the 3 years of observation, from 9–15 February in 2007 to 17–23 January and 18–24 December in 2008, and 1–7 January in 2009.

Figure 2.

 Mean female Aedes aegypti collection per trap per day at 11 sites in Cairns Queensland from September 2006–April 2009, with concurrent temperature and humidity data (*collection lasted for 2 weeks because of non-collection because of dengue outbreak response).

Regression analysis: including longer- and shorter-term factors

Only three of all the meteorological variables in the analysis were included in the regression models: mean minimum temperature of lag 6 months, mean daytime temperature of lag 4 months and mean relative humidity of lag 2 weeks (Table 1). Other variables were not included because of a lack of significant association with A. aegypti abundance. A decrease in monthly mean minimum temperature 6 months prior to mosquito collection week was the most significant entomological indicator for the rise in daily abundance of female A. aegypti (i.e. meanfemday) as collected from BG-sentinel traps. This is followed by a decrease in monthly mean daytime temperature 4 months prior to mosquito collection week. Mean relative humidity from previous 14 days also seems to have an effect on the increase of meanfemday, but to a lesser extent and in a different direction compared to the temperature factors.

Table 1.   Longer- and shorter-term meteorological factors that have (the most) significant relationship with mean number of female Aedes aegypti per day (Model 1)
VariableCoefficientSECI95r2N
Meanfemday
 MINTEMP lag 6 month−0.0360.003−0.043, −0.0290.638132
 TEMP lag 4 month−0.0110.004−0.019, −0.003  
 RH lag 2 week0.0040.0010.001, 0.007  
 Constant1.0530.1110.834, 1.273  
Model 1: Meanfemday = 1.053 + (−0.036)*MINTEMP lag 6 month + (−0.011)*TEMP lag 4 month + 0.004*RH lag 2 week

Regression analysis: including shorter-term factors only

For the analysis of short-term meteorological factors, only mean daytime temperature of 0 week (i.e. current week) and mean relative humidity of lag 2 weeks significantly related to meanfemday (Table 2).

Table 2.   Significant shorter-term meteorological factors that have significant relationships with mean number of female Aedes aegypti per day (Model 2)
VariableCoefficientSECI95r2N
  1. *Heteroscedasticity was detected in this model. Hence, regression with robust standard errors was performed.

Meanfemday
 TEMP 0 week0.040.0030.034, 0.0460.610127
 RH lag 2 week0.0050.0010.003, 0.007  
 Constant−1.080.077−1.232, −0.928  
*Model 2: Meanfemday = −1.08 + 0.04*TEMP 0 week + 0.005* RH lag 2 week

Model 2 validation

A high correlation coefficient (r2 = 0.762) was observed between actual and predicted meanfemday albeit with different scales – lower in the actual values and higher in the predicted values (Figure 3).

Figure 3.

 Values of predicted mean female Aedes aegypti per trap per day (meanfemday) as calculated from Model 2 and actual meanfemday from 23 April until 26 November 2009.

Of 33 data sets, 31 are within upper and lower limits of agreements, indicating a strong concordance between the predicted and actual mean female A. aegypti per trap per day (Figure 4). The limits of agreement were calculated from ± 1.96s where d is mean of the difference between each pair of predicted and actual values, whilst s is standard deviation of the difference between these pairs (Bland & Altman 1986).

Figure 4.

 Bland–Altman agreement plot shows the difference between each pair of predicted and actual meanfemday (Y axis) plotted against the average of each pair predicted and actual meanfemday (X axis). Values between the two dashed lines are within the limits of agreement.

Discussion

Model 1 (Table 1) demonstrates the seasonal influences on A. aegypti abundance and describes the general wet season peak in activity. It reveals significant relationships between mean female A. aegypti and lagged average daytime temperature and minimum temperature of 4 and 6 months, respectively. The coefficients for these variables were negative, indicating an increase in vector abundance could be forecasted by a decrease in monthly temperature 4 and 6 months previously. This result does not add significantly to our understanding of A. aegypti population flux, in that is merely the broad seasonality of these mosquitoes in the tropics; namely that they are more abundant in times of higher humidity and temperature (i.e. the wet season).

Model 2 (Table 2) explains more immediate effects of weather on vector abundance and for this reason was chosen for validation testing. This model was found to be quite sensitive in predicting the sharp reduction in meanfemday from end of April until mid-May 2009 and the sharp increase in meanfemday from mid-November until end of November 2009. The model also considerably explains the trends of meanfemday from mid-August until mid-October 2009. Nevertheless, this model failed to describe the fluctuations of meanfemday in several instances, for example from end of June until early July 2009 when vector abundance became very low (Figure 3).

Based on Model 2 validation, it is apparent that weather-based predictive modelling can be used to forecast the short-term increases and decreases in vector abundance (Figure 3). This would allow vector control measures to be initiated before the sharp increase happens. Daily weather forecasts can be obtained from BoM for 7 days in advance but not beyond that (Climate Information Services, Bureau of Meteorology, personal communication 4 January 2010), and daily relative humidity data over the preceding 14 d are readily obtainable on-line (http://www.bom.gov.au), meaning that calculations of vector abundance for the coming week can be made using Model 2.

At present, the trigger for vector control operations (in the absence of dengue transmission) is five female A. aegypti per week per trap for three consecutive weeks; a threshold that is arbitrarily derived. Vector control and source reduction are then undertaken at that monitoring site and its two or three surrounding city blocks (Brian Montgomery, TPHU Queensland Health, personal communication 15 June 2009). An ideal vector monitoring system should have adequate efficacy to prospectively detect an increase in vector abundance and to treat containers before large number of adults emerge. Simultaneously, the system also needs to provide minimal false predictions (Runge-Ranzinger et al. 2008). Based on Model 2 (Table 2), vector control measures could be intensified prior to the detected vector abundance increase in traps.

Pre-emptive vector control could be targeted towards premises which have been found in the past to harbour a considerable number of potential breeding habitats for mosquitoes. Mass deployment of lethal ovitraps (biodegradable or otherwise) (Rapley et al. 2009; Ritchie et al. 2009) in areas of historically higher dengue risk could also be employed. Media campaigns could be started earlier at this stage to inform the public that dengue vectors are on the rise, and steps to minimize the likelihood of a dengue outbreak might be taken. Therefore, predictive tools such as described in this study could partly inform decision-making process in determining the time to initiate these control measures.

A similar analysis of meteorological factors and vector abundance was undertaken in Argentina (Estallo et al. 2008). A combination of 13 variables of meteorological data and satellite-derived indicators was used to construct predictive models of House – and Breteau Indices. Such a combination was found to be of potential use in predicting these larval indices particularly when time lags of the 13 variables were included into the analysis. Specifically, it was found that a 15-day time lag for humidity was significantly associated with vector abundance (Estallo et al. 2008); similar to the significant 14-day lag reported here (Tables 1, 2). When actual field data were compared to the predictive models, scale differences were evident between these two series but a significant agreement in temporal fluctuation of the larval indices was displayed (Estallo et al. 2008); a phenomenon also observed in this study with adult abundance (Figures 3 and 4).

Another study showed that rainfall and temperature began to rise 3 and 7 weeks respectively before entomological indices started to escalate (Dibo et al. 2008). These previous studies support the results of our analysis that suggest temperature and relative humidity as the most significant meteorological factors in determining dengue vector abundance. Further, we have attempted to scrutinize the relationship into a shorter time period (Table 2) and found that mean relative humidity from the previous 14 days has a significant relationship with mean number of female A. aegypti per day.

Importantly, no rainfall variables formed part of the final predictive models presented here. Rainfall was found to be highly collinear with temperature and humidity variables and was not as strongly predictive of A. aegypti abundance. While rainfall is clearly required to fill breeding containers and to provide humidity, variation in rainfall is not as strongly related to variation in vector abundance as other variables. This may be in part attributed to the importance of manually filled containers (e.g. pot plant saucers) in local A. aegypti population dynamics (Williams et al. 2008).

Despite the ability to predict fluctuations in vector abundance, the indistinct link between this and dengue risk may limit the utility of predictive models. In one study, it was found that both adult and egg count vector indices had good correlation with dengue incidence rate (Dibo et al. 2008), thus proving them to be effective in describing the disease within period of active transmission. Conversely, when the transmission had subsided, the increase in these entomological indices could no longer be associated with the number of dengue cases despite their agreement with rainfall and temperature variables (Dibo et al. 2008). Similar results were shown with the Breteau index (BI). In one study performed in Trinidad, BI coincided with rainfall patterns and dengue incidence in 2002 and 2003. However, while BI remained high in wet season 2004, the number of dengue cases had been lower than the previous 2 years (Chadee et al. 2007). High BIs involving large numbers of low mosquito productivity container types can provide misleading indicators for dengue risk (Kay et al. 1984). Based on these findings, the ability of entomological indices and meteorological factors as measures of dengue fever risks could be doubted, especially in quiescent periods of dengue transmission. Other factors including herd immunity and introduction of a new viral strain or serotype could be considered to justify this phenomenon (Loh & Song 2001; Brunkard et al. 2007; Chadee et al. 2007).

In a Taiwan study, BI was found to be inversely related to dengue fever incidence even after 1–6 months lag were imposed on the BI to explain the relationship (Wu et al. 2007). This is supported by several other studies that also demonstrate that entomological indices assessing vector density have failed to be associated with dengue incidence and/or weather variability (Focks & Chadee 1997; Bangs et al. 2006; Chadee et al. 2007).

Our findings here, together with those of previous research, provide an indication that readily available meteorological data can be used to predict changes in vector abundance, but not necessarily dengue risk. Moreover, the importance of meteorological data in describing mosquito abundance in short temporal scales has not been limited to dengue vectors, but also other species of mosquitoes in Australia, for example Aedes vigilax and Aedes camptorhynchus– both vectors of Ross River virus (Kokkinn et al. 2009; Williams et al. 2009; Yang et al. 2009).

In this study, the linear regression models presented only explain 61% and 64% of the variation in mean number of female A. aegypti per day. Endogenous factors such as human population growth rate and carrying capacity (i.e. breeding container density) for dengue vectors (Otero & Solari 2005; Yang et al. 2008) might have to be considered to develop further understanding on how other factors contribute to the fluctuation and magnitude of vector abundance.

The validation of the shorter-term factors model (Model 2) was successful, albeit with a slight scale difference between predicted and actual values. On average, for each of the 33 validation measurements, the actual number of female A. aegypti was 0.18× the predicted number. This difference might be associated with the intensity and efficacy of dengue responses performed by the Dengue Action Response Team (DART) during the recent dengue outbreak in 2008/2009 which caused the meanfemday values in 2009 to be lower than previous years. The outbreak, which lasted from November 2008, had witnessed a large-scale vector control operation until it was declared over in September 2009. Within this period, more than 518 000 possible mosquito breeding sites were treated with residual larvicides, and more than 8300 interior residual sprayings were performed by Queensland Health (Queensland Health 2009).

Our findings suggest that minimum temperature and average daytime temperature are the most significant factors to be associated with vector abundance in the longer- and shorter-term, whilst relative humidity has a significant effect in shorter-term analysis. Rainfall was not a significant determinant of changes in A. aegypti adult abundance. Such findings could contribute to the forecasting of vector abundance peaks and subsequently guide pre-emptive vector control operations. These measures would help in minimizing the spread of dengue infection and prevent the occurrence of dengue outbreak in the community provided they are applied in a sustainable and timely manner. Clearly, meteorological variables can be used prospectively to predict short-term changes in dengue vector abundance, using Model 2 described previously. Such predictions can be used to guide the application of pre-emptive dengue vector control, and thereby enhance the management of this disease.

Acknowledgements

The authors acknowledge the Malaysian Ministry for Higher Education and Universiti Kebangsaan Malaysia for financial support to Aishah H. Azil. Brian Montgomery and Karel Van Horck (Queensland Health), Michelle Larkman (JCU) and members of Dengue Action Response Team, TPHU Queensland Health are thanked for technical and logistical support for this study. Tamika Tihema from Queensland Climate Services Centre, Bureau of Meteorology is acknowledged for providing the weather data.

Ancillary