An open source tool to reduce geolocation uncertainty in GEDI data

NASA's GEDI mission, launched in December 2018, is the first spaceborne lidar optimised to measure forest structure. It uses a full waveform lidar with ~22 m diameter footprints to measure samples of vertical structure. These samples are used to derive a number of data products, including aboveground biomass density, canopy cover, ground elevation and leaf area index profiles. Footprint level validation of these products requires accurate collocation between GEDI and data collocated from ground-based or airborne instruments. GEDI's expected geolocation accuracy of 7-8 m (1 sigma) can be improved by maximising the correlation between GEDI waveforms and simulations derived from discrete-return airborne laser scanning (ALS) data. This uses ALS data to simulate GEDI with a range of potential horizontal and vertical geolocation offsets. The correlation between real and simulated data is calculated for each offset and the maximum taken as the true offset. The method was made more computationally efficient by adding a simplex optimiser to find the maximum correlation without having to sample a full grid of potential offsets.

To test the ability of this tool to accurately collocate GEDI and ALS data, simulations of GEDI data with different geolocation errors (both whole track errors and shot-to-shot jitter errors) were made over a wide range of biomes. The shape of the correlation surfaces were analysed in order to determine whether there was always a clear global maximum, and how close a simplex optimiser must be started to this to avoid finding a local maximum. Results showed that every area examined had a clear global maximum, with a biome dependent separation of at least 6 m horizontally and 2 m vertically between the global maximum and the nearest saddle point to a local maximum. The much shorter tolerance in the vertical direction is not surprising given the shorter vertical length of the laser pulse (3.8 m) compared to the horizontal width (22 m). These correlation distances were used in an initial full-grid search to determine a starting point and a simplex correlation optimiser started from there. This method successfully collocated simulated GEDI and ALS data with less than 1 m error, as long as the area included above a minimum number GEDI footprints.