Monitoring Deforestation Using Satellite Radar Imagery

Forests play a major role in our fragile ecosystem. When planted they act as a carbon sink, but when cut down or burnt they act as a carbon source. In light of the universal awareness of the reality and consequences of global climate change, significant efforts are undertaken to preserve forests and reforest areas that have experienced large scale deforestation. Due to large scale coverage and regular revisit times, satellite remote sensing is the most promising candidate for any forest cover monitoring system.

The aim of the project is thus to develop and implement efficient new algorithms for analysing sequential images (such as optical or radar satellite) to detect temporal changes and anomalies. Sequential image data may include both small and large-scale spatial and temporal trends combined with abrupt changes, as a result of genuine changes in the underlying data structure, with associated observational noise overlaying this structure. Such data arise in many application areas, including not only monitoring of deforestation, but also detecting compositional changes in pipelines and safety monitoring of objects to identify corrosion/cracks.

With new technology, the amount and resolution of such data has exploded, with associated big data challenges. The primary focus is the detection changepoints in the small-scale composition of the images over time, taking into additional account factors such as smooth spatial changes and temporal cyclical trends, as well as observational noise.