Quantitative spatio-temporal sequence discovery

Quantitative spatio-temporal (ST) sequences represent chains of ST event types in a ST framework. For example, it has been widely observed that deforestation in peat-land forests are often followed by forest fires, in close proximity of space and time. Such patterns between deforestation and forest fires are attributed to the physical phenomena of degradation of soil due to logging, which reduces the moisture in the soil, making the underneath peat reserves more susceptible to fire. Given an array of multiple real-valued signals representing distinct features of the domain, a user-defined interest measure, and a spatio-temporal neighborhood relation, Quantitative ST sequence discovery reports all sequential patterns with interestingness above a user-defined threshold. Current literature primarily focuses on boolean event types with instances represented as points in space and time. However, in many application domains including, forestry, the input datasets are raster, and the events span arbitrary time periods. In this project, we attempt to formalize the quantitative ST sequence discovery problem that deals with raster data inputs while accounting for events with arbitrary time intervals.