The prevalence of dengue fever poses a global health issue, with the World Health Organization (WHO) estimating that half the world’s population are now at risk from the mosquito-borne viral infection. The WHO says it has seen an eight fold increase in reported cases over the last two decades, from 505,430 cases in 2000, to over 2.4 million in 2010, and 4.2 million in 2019.
The key to controlling dengue is costly surveillance and control of the mosquitos that spread the disease, and forecasting outbreaks can contribute greatly to the distribution of scarce resources. Better forecasting equals better management and, the hope is, fewer cases in the long term. Predicting outbreaks also allows more targeted support to help local communities reduce mosquitos through a range of simple measures such as improving water storage and waste management, as well as personal protection such as mosquito coils.
Funded by the UK Space Agency and developed by a consortium led by HR Wallingford, the state-of-the-art Dengue Model forecasting Satellite-based System (D-MOSS) forecasts outbreaks six months in advance and is the first functional system that uses EO data, in-situ observations and seasonal climate forecasts to issue forecasts on a regular basis. A prototype D-MOSS system is operating in Vietnam and, owing to the success of the trial there, it is now being implemented in Malaysia and Sri Lanka, and rollouts are planned in four other Asian countries.
The D-MOSS concept
Based on the latest epidemiological research, hydro-meteorology and informatics, D-MOSS is designed to overcome the practical challenges of forecasting mosquito-borne diseases. For instance, it takes into account how temperature influences mosquito development; reproduction rates; mosquito distribution; how rainfall creates or destroys breeding sites; land use; and mosquito control measures.
By collecting 20-year histories of satellite-based EOs from services including NASA DAAC, Copernicus and NOAA CFSv2, D-MOSS is able to establish the relationship between these factors and dengue. Gaps in the satellite data are filled using interpolation and comparison to in-situ observations from ground stations, which create a ‘big data’ repository of hydro-meteorological, land cover and population variables. These include maximum and minimum temperature, precipitation, soil moisture, urban and peri-urban area percentages, and population totals. Designed using an advanced information architecture based around the spatio-temporal structure of each dataset, the system calculates data for every Vietnamese province by harmonising the datasets into a timeseries set of polygons taken from the province boundaries, referred to as a MultiPolygonSeries.
At the start of the forecasting process, a water availability model is used to generate a set of parameters representing the amount of water present in the earth surface environment (such as total runoff and evapotranspiration). After this, the relationships between the hydro-meteorological variables and the numbers of cases of dengue fever are created by a statistical superensemble, generated by Bayesian model averaging. This process is repeated monthly or weekly as new data arrives from satellites and as new cases of dengue fever are reported.
To produce forecasts, D-MOSS incorporates the UK Met Office GloSea5 six-month seasonal forecast of hydro-meteorological variables to predict outbreaks. Forecasts are displayed on a private website using interactive maps, charts and tables and, to make the results relevant to policy makers and practitioners, they are presented as the probability that cases will exceed thresholds set by organisations such as the WHO.