1. PollutantsEventsDiseases
Analysis Scope: A set of pollutants pol1,…,polk, diseases d1,…,dm and disaster events e1,…,er
Given Data (or need to be created during the course of the project):
A spatio-temporal grid of pollution observations P pol1,x,y,t ,…, polk,x,y,t where poli,x,y,t denotes the concentration of the i-th pollutant at grid-cell (x,y,t) for an observation period
A set of disease outbreak instances D described in the form: <disease, longitude, latitude, time>
A set of disaster events E in the form: <event-type, longitude, latitude, start-time, end-time>
Project Outcomes:
A spatial pollution base line for pollutants pol1,…,polm for each spatial grid cell (x,y):
Pol(j,x,y)= Raw average pollutant polj concentration in cells (x,y,1),…,(x,y,tmax)
Pol’(j,x,y):= Baseline average of pollutant polj; same as Pol(j,x,y) but does not use cells (i,j,t’) in which disaster events occurred in average/… computations; similarly, we can compute base line standard deviations.
Event-based (es for s=1,…,r) pollutant spike summaries that are obtained by comparing concentrations of particular pollutants associated with a particular event type (e.g. fire) with the pollutant’s baseline.
Conceptual/geospatial models that associate pollutant concentrations with the occurrence of particular diseases; a “naïve” approach to get those could be: compute disease densities in each pollution grid-cell; then learn models mi using each grid-cell as training examples that map pollutant concentrations into densities/probabilities of the k-th disease (for i=1,…,m).

2. Assumptions and Discussion
Conceptual models are functions that map pollutant concentrations into disease probabilities/densities; e.g. a model might map benzene and chlorine concentrations into lung disease probabilities.
Disease outbreak delays represent a challenge for creating conceptual models.
As conceptual models are learnt from spatial pollution grids, conceptual models can be used to compute regional disease probability maps from pollutant concentration grids.
The previous slide suggests to compute baselines by excluding observation that are associated with disaster events (e.g. oil spill); an alternative approach could be to apply an outlier removal technique to the concentrations of a particular pollutant, and then to compute the baseline based on the “surviving” observations.
The current project description also says something about “identifying pollution sources”, which is not considered in the previous slide that centers on obtaining pollutant concentrations and on computing baselines, event spike summaries and conceptual models from those. In general, I believe this is a very challenging task.
Remark: Project outcomes can be computed by changing grid granularities.