An alternative AMSR2 vegetation optical depth for monitoring vegetation at large scales

Vegetation optical depth (VOD) retrieved from microwave observations has been found to be useful to monitor the dynamics of the vegetation features at global scale. Particularly, many applications could be developed in several fields of research (ecology, water and carbon cycle, etc.) from VOD products retrieved from the SMOS and SMAP observations at L-band, and from the combined AMSR-E (2002-2011)/AMSR2 (2012-present) observations at C-and X-bands. One of the main difficulties in retrieving VOD is that the microwave observations are sensitive to both the soil (mainly soil moisture) and vegetation (mostly VOD) features. The AMSR-E/2 sensors provide only mono-angular observations at two polarizations. These dual-channel observations may be strongly correlated so that retrieving SM and VOD simultaneously can be an ill-posed problem. Here, to overcome this problem, we proposed and evaluated a new retrieval approach from AMSR2 observations at X-band to produce a new X-VOD product. The X-VOD retrievals were based on the inversion of the X-MEB model, an extension of the L-MEB model (L-band microwave emission of the biosphere) to the X-band. The main originality in comparison to previous algorithms is that (i) only VOD was retrieved while SM was estimated from a reanalysis data set (ERA5-Land); (ii) model inversion was based on an innovative approach to initialize the cost function; and (iii) new values for the soil and vegetation X-MEB model parameters were calibrated. To evaluate the methodology, we performed the VOD retrievals over the whole African continent over 2014-2016, including a dry (2015) and a wet (2014) year. In a first step, we calibrated a set of three parameters: effective scattering albedo (omega), soil roughness (H-R) and VOD first guess (VODini). Several datasets of vegetation indices as Above-Ground Biomass (AGB), Leaf Area Index (LAI) and Normalized Difference Vegetation Index (NDVI) were chosen as reference data to optimize these model parameters. Globally-constant values (omega = 0.06 and H-R = 0.6) were found to achieve high spatial and temporal correlations between retrieved X-VOD and the reference vegetation parameters. Comparison with other X-VOD products suggested IB X-VOD had competitive advantages in terms of both spatial and temporal performances. In particular, spatial correlation with three biomass datasets was found to be higher than for previous X-VOD products (R-2 similar to 0.76-0.83) and temporal correlation with LAI or NDVI showed obvious improvements, especially in dense vegetation.