Classification & Segmentation
We develop and apply techniques to classify data sets in order to better understand cryospheric processes. This includes a mix of both unsupervised and Artificial Intelligence (AI) techniques in order to contrast the benefits and limitations of both. Some of the projects we are actively working on include iceberg mapping in glacier fjords, classification of features on glaciers surfaces (e.g. hydrology, debris, crevasses) and mapping of snow surface features. We are also interested in multi-modal, multi-resolution data fusion and developing methods to combine data sets of various forms to improve traditional classification/segmentation methods.
3D Data Extraction
3D data products are critical for understand glacier mass loss, freshwater fluxes into the ocean, volume transfer, sea level changes amongst others. At a large scale, altimeters have been used to extract coarse 3D data sets over glaciers and ice sheets but these cannot be used to understand processes such as iceberg melt rates, iceberg calving and surface melt rates. We therefore develop methods to extract high resolution 3D data products from different sensors and thus aim to fill gaps left by satellite sensors. For example, we are actively developing methods to extract 3D data sets from Synthetic Aperture Radar (SAR) imagery using techniques such as ‘Shape-from-Shading’. We also aim to develop close-range remote sensing systems to further fill 3D gaps. Our approach is therefore multi-model, multi-resolution and we are also seeking methods to fuse these data sets.
Time Series Analysis & Forecasting
We have have over 50 years’ of Earth Observation data from satellites plus a diverse range of glaciological measurements collected by international teams. We aim to bring these data sets together to analyse time series of changes across the Arctic and aim to use this build forecasting models. For example, we are currently working on building sea ice forecasting models to predict future changes in sea ice coverage to analyse its effect on sustainable shipping practices. These techniques utilise deep learning methodologies to combine multi-model data sets from various sources. We also place a strong emphasis on quantifying the uncertainty of our measurements and reporting these alongside our data products. We are also interested in developing techniques to combine uncertainties and ensure uncertainties are calculated in a suitable way when fusing multi-model data products.
Computer Vision
We are interesting in applying computer vision techniques to extract new data prodicts from remote sensing imagery, particularly from satellites. For example, we are developing methods to analysis satellite images using texture analysis techniques such as the Gray Level Co-occurence Matrix (GLCM). Further, we also aim to apply new computer algorithms to extract unique and novel data products from image data such as more accurate velocity retrievals, 3D data sets, statistical features amongst others.
Research Page: William D. Harcourt 