spatiotemporal version of the convolution model to infer malaria infection incidence maps from clinical incidence data and prevalence surveys

[smwindecker]

From @goldingn in inferepiparams thread:

Use a space-time GP (e.g. with the banded SPDE approximation I have a prototype of) + covariate effects to model (log) infection incidence at pixel level and monthly. Use a PCR positivity curve to convolve this forward to prevalence, and use in a (beta)binomial observation model for point surveys (single lat long and date). Aggregate infection incidence spatially to clinic catchment areas, then convolve with a symptom onset delay distribution (and weight according to detection rate) to define a Poisson/NB count model over clinical incidence time series for those facilities. If there is enough prevalence data, the model could learn changes in detection effort (RDT availability, detection of asymptomatics due up background fever).

It could be an extension or sister package but doesn't fit in lowerGPreff/existing suite of pacakges.