Automated BRDF Correction of Multiple Adjacent Imaging Spectroscopy Flightlines in Complex Landscapes

Abstract:

Bi-directional reflectance distribution function (BRDF) effects are a persistent issue for the analysis of vegetation in airborne imaging spectroscopy data, especially when the goal is to mosaic analyses from multiple adjacent flightlines (collectively, a flight box) or for change detection. In addition, with the advent of large airborne imaging efforts such as NASA’s Arctic Boreal Vulnerability Experiment (ABoVE), India and California campaigns and the annual flights by the US National Ecological Observatory Network (NEON), there is increasing need to develop methods that are automatable across large numbers of images with diverse land cover. We developed an automated method to correct for BRDF effects using a series of nine long (150-400 km) Airborne Visible/Infrared Imaging Spectrometer (AVIRIS-Classic) flightlines flown on May 22nd, 2013 over Southern California. In these flightlines, rough terrain and diverse land cover created significant BRDF effects in which the traditional single-image BRDF corrections did not compensate for between-image brightness effects. Using the widely employed kernel-driven semi-empirical Ross-Li BRDF correction approach, we sampled from each image in the flight box and used the pooled values to calculate correction coefficients for the entire group. To account for variable anisotropic scattering properties of different vegetation types, unique correction coefficients were calculated along and interpolated from an 18-bin NDVI gradient. Comparison of root mean square error (RMSE) and mean absolute deviation (MAD) of pixels in overlapping areas of images show that correction algorithms built for the entire flight box performed better than corrections built for single images. To demonstrate the broad applicability of the method, we applied our approach in a variety of environments with different sensors, including AVIRIS-Next Generation imagery from India and the Arctic, as well as NEON imagery from Wisconsin. For AVIRIS-C and AVIRIS-NG, RMSE improved 0.1-1.8% while MAD improved 0.3-2.2%. With the NEON imagery, there was essentially no improvement (nor worsening), likely due to minimal initial BRDF effects. Finally, we demonstrate the implications of our correction approach using derived indices and maps of vegetation traits.