Developing novel products to solve real world issues using EO data - e.g.- Land Use, Land Use Change and Forestry (LULUCF) applications to improve GHG inventories in the UK- A fire risk index using a probabilistic approach with TIGGE global ensemble forecast data and MODIS data- Virtual Sensor concept merging observations from Sentinel-3 (OLCI), MODIS, Sentinel-2 (MSI) and PlanetScope. Retrieving vegetation biophysical parameters using both physical and machine learning approaches (e.g. radiative transfer models and Gaussian processes)

As part of the EU-funded MELODIES project I design and developed a Land Cover Change mapping methodology for the UK to improve GHG emission inventories using an optimal estimation framework using MODIS data and a Support Vector Machine (SVM) classifier.

As part of the GREENCYCLES II initial training network I developed a long-term (1981-2011) burnt area map database at 0.05º spatial resolution during a 16-day period. The burnt area product initially will be used to create a fire regime global synthesis map and eventually as well to improve estimates of pyrogenic emissions of greenhouse gases, to analyse fire-climate relationships, and to assess the performance of process-based fire modules of dynamic global vegetation models (DGVMs).The main input of the burnt area identification algorithm is the Advanced Very High Resolution Radiometer (AVHRR) long-term data record (LTDR) which comprises a daily global atmospherically-corrected surface reflectance dataset at 0.05º spatial resolution. The AVHRR-LTDR is available for the period 1981-1999, in order to complete the surface reflectance time series until 2008, the Moderate Resolution Imaging Spectroradiometer (MODIS) sensor will be used. However the use of another sensor implies, either a sensor intercalibration or a band conversion to a common wavelength, a commonly used methodology is the narrow to broadband transformation. The latter option was implemented using the narrowband to broadband conversion formulae for AVHRR developed by Liang (2001), this will allow to integrate MODIS and AVHRR data in three spectral broadband ranges, namely the solar spectrum (400-3000nm), the visible (400-700nm) and the near- and shortwave infrared (700-3000nm). All daily AVHRR acquisitions were converted to broadband for the chosen prototype year, 1998.

- Development of a multi-sensor broadband Albedo product within the ESA GlobAlbedo project using MERIS, VGT and MODIS data using an optimal estimation framework.- Operational generation of the 15-year albedo global time series in a HPC Linux cluster at UCL. Scientific programming in Python and Java.