Estimativa de biomassa em fitofisionomias do Cerrado usando dados do sensor Hyperion e regressão por mínimos quadrados parciais

Estimation of aboveground biomass (AGB) is challenging in large and complex areas such as the Brazilian Cerrado, which requires improved approaches to achieve the necessary accuracy. The development and launch of orbital hyperspectral sensors provide an opportunity to estimate the AGB using a number of metrics derived from these images. As a preparatory study for future hyperspectral missions, the AGB was estimated from the Hyperion images obtained by the Earth Observing-1 (EO-1) platform over the Ecological Station of Águas Emendadas, in central Brazil. We tested four groups of attributes with the partial least squares regression (PLSR): 146 spectral bands; 22 vegetation indices; 10,585 band ratios; and 24 absorption band parameters. We developed specific AGB/PLSR models for each group of attributes and tested a general model. The results showed a better performance with the use of all groups of spectral attributes in the general model (R2 = 0.66, RMSE = 6.60 t.ha-1) than with the use of each set of attributes. These values were comparable to those observed from the AGB-derived band ratio model (R2 = 0.65, RMSE = 6.63 t.ha-1). The Hyperion data allowed combination of different hyperspectral attributes in the AGB modeling. This may contribute for developing future works in the Cerrado ecosystem with the next generation of orbital hyperspectral sensors such as the Environmental Mapping and Analysis (EnMAP) and Hyperspectral Infrared Imager (HyspIRI), with much better signal-to-noise and larger swath width than the Hyperion.