Application of satellite precipitation data to analyse and model arbovirus activity in the tropics

Abstract

Background: Murray Valley encephalitis virus (MVEV) is a mosquito-borne Flavivirus (Flaviviridae: Flavivirus) which isclosely related to Japanese encephalitis virus, West Nile virus and St. Louis encephalitis virus. MVEV is enzootic innorthern Australia and Papua New Guinea and epizootic in other parts of Australia. Activity of MVEV in WesternAustralia (WA) is monitored by detection of seroconversions in flocks of sentinel chickens at selected sample sitesthroughout WA.Rainfall is a major environmental factor influencing MVEV activity. Utilising data on rainfall and seroconversions,statistical relationships between MVEV occurrence and rainfall can be determined. These relationships can be usedto predict MVEV activity which, in turn, provides the general public with important information about diseasetransmission risk. Since ground measurements of rainfall are sparse and irregularly distributed, especially in northWA where rainfall is spatially and temporally highly variable, alternative data sources such as remote sensing (RS)data represent an attractive alternative to ground measurements. However, a number of competing alternatives areavailable and careful evaluation is essential to determine the most appropriate product for a given problem.Results: The Tropical Rainfall Measurement Mission (TRMM) Multi-satellite Precipitation Analysis (TMPA) 3B42product was chosen from a range of RS rainfall products to develop rainfall-based predictor variables and buildlogistic regression models for the prediction of MVEV activity in the Kimberley and Pilbara regions of WA. Twomodels employing monthly time-lagged rainfall variables showed the strongest discriminatory ability of 0.74 and0.80 as measured by the Receiver Operating Characteristics area under the curve (ROC AUC).Conclusions: TMPA data provide a state-of-the-art data source for the development of rainfall-based predictivemodels for Flavivirus activity in tropical WA. Compared to ground measurements these data have the advantage ofbeing collected spatially regularly, irrespective of remoteness. We found that increases in monthly rainfall andmonthly number of days above average rainfall increased the risk of MVEV activity in the Pilbara at a time-lag oftwo months. Increases in monthly rainfall and monthly number of days above average rainfall increased the risk ofMVEV activity in the Kimberley at a lag of three months.In