When forecasting epidemics, the main interests lie in understanding the determinants of transmission and predicting who is likely to become infected next. However, for vector-borne diseases, data availability and alteration can constitute an obstacle to doing so: climate change and globalized trade contribute to the expansion of vector habitats to different territories and hence the distribution of many diseases. As a consequence, in the face of a rapidly changing environmental and ecological climatic conditions, previously well-fitted models might become obsolete soon. The demand for precise forecast and prediction of the spread of a disease requires a model that is flexible with respect to the availability of vector data, unobserved random effects and only partially observed data for diseases incidence. Thus, we introduce a combination of a mechanistic SIR model with principled data-based methods from geostatistics. We allow flexibility by replacing a parameter of a continuous-time mechanistic model with a random effect, that is assumed to stem from a spatial Gaussian process. By employing Bayesian inference techniques, we identify points in space where transmission (as opposed to simply incidence) is unusually high or low compared to a national average. We explore how well the spatial random effect can be recovered within a mechanistic model and only partially observed outbreak data available. To this end, we extended the Python probabilistic programming library PyMC3 with our own sampler to effectively impute missing infection and removal time data.