In low burden areas of the world, surveillance data (number of clinical cases reporting to health centres) is often a more informative measurement of malaria burden than parasite rate surveys because a large proportion of the population is being passively sampled continuously.
The ROAD-MAP team have collected a large database of routine surveillance data and their associated geographic geometries. However, statistical methods based on geopositioned point data cannot be used with these areal data and traditional models for areal data cannot be used to create estimates at high-resolution pixels.
We are developing statistical methods that can handle these areal data to estimate high-resolution maps of Plasmodium falciparum incidence and and Plasmodium vivax globally. This includes new approaches for mapping in low-burden settings and utilising varied types of response data, thus necessitating new modelling approaches.
# Model overview
We are developing disaggregation models that can relate areal malaria incidence data to pixel level environmental and human covariates.
To combine information from parasite rate surveys with these disaggregation models we are using maching learning models.
Finally, to incorporate fine temporal pattern into our estimates we are calibrating out predicted malaria risk surfaces to estimates from national time-series models.