Publications

Santi, E; Paloscia, S; Pampaloni, P; Pettinato, S; Nomaki, T; Seki, M; Sekiya, K; Maeda, T (2017). Vegetation Water Content Retrieval by Means of Multifrequency Microwave Acquisitions From AMSR2. IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, 10(9), 3861-3873.

Abstract
In this study, two retrieval algorithms for estimating the water content of vegetation (VWC) in the range 0-8 kg/m(2) from multifrequency microwave radiometric data have been implemented, with the purpose of contributing to the development of an all-weather VWC product for the satellite AMSR2 radiometer of the JAXA (Japan Aerospace Exploration Agency) GCOM-W mission. The first algorithm estimates VWC through a semi-empirical combination of the polarization index (PI) acquired at X and Ku bands, while the second one endeavors to improve the retrieval accuracy adding C-band data and using an artificial neural network (ANN) method. The sensitivity to vegetation biomass of multifrequency PIs at various frequencies, which was already pointed out in the previous literature, has been first further evaluated with the support of the well-known tau-omega solution of the radiative transfer model and experimental data. Two years of AMSR2 data collected on a wide portion of Africa, which includes a large variety of vegetation types and biomasses, have been considered for implementing and testing both algorithms. VWC reference data with the same temporal and spatial coverage of AMSR2, needed for validating the algorithm outputs, have been derived from SPOT4 normalized difference vegetation index (NDVI), downsampled to the AMSR2 ground resolution. The test results provided a correlation coefficient R > 0.88 with root mean square error (RMSE) < 1.4 kg/m(2) for the semi-empirical algorithm, and R = 0.98 with RMSE < 0.5 kg/m(2) for ANN algorithm, thus demonstrating that, although both approaches are able to estimate VWC, the ANN algorithm is able to obtain better results. An independent validation of the two algorithms was then carried out on the entire Australian continent, considering 4 periods of 16 days each in different seasons of 2013. In this case, the algorithm outputs have been compared with VWC derived from MODIS 16 days averaged NDVI. The independent validation, as expected, showed slightly worst results, with R = 0.82 and RMSE around 1.5 kg/m(2) for the semi-empirical algorithm, and R = 0.88 and RMSE around 0.86 kg/m(2) for the ANN. This study demonstrated that microwave data from AMSR2 and polarization indices can be legitimately used to produce vegetation maps on a global scale by separating several levels of biomass, without any need of further information from other sensors.

DOI:
10.1109/JSTARS.2017.2703629

ISSN:
1939-1404