Re-search

Conventional ocean colour inversion models (that which derive in-water constituent concentrations from above surface reflectance signals) suffer from non-unique solutions and equifinality problems, i.e., the different combinations of different Bio-Geo-Chemical parameters (BGC: TSS, DOC, TChla concentations) can result in the same reflectance spectra. We propose a unique way of addressing this issue by simultaneously retrieving the inherent optical properties of the water parcels (including the respective backscattering and absorption of particulate and dissolved substances) along with these BGC values using a deep learning framework trained on augmented spectral library referenced to in-situ measurements.

Project info:  CSIRO AI4Missions,  NSW Marine Estate Management Strategy

Existing flood inundation models suffer from large computational complexities and while deep learning approaches, although efficient, require large volumes of dynamic catchment datasets including discharge and flow velocity, that are near impossible to obtain in flood scenarios. We propose a novel conceptual flood routing and a Height Above Nearest Drainage (HAND) based inundation framework for mapping flood extents across large-scale data-scarce regions with significant reduction in computational complexities.

News coverage:  Monash Lens, Monash University, YouTube