Abstract:
Earth observation via optical-based remote sensing is one of the effective solutions to cover the large swath and to deliver the very highresolution dataset at the different wavelengths. But the applicability of optical imaging is limited by daytime only and adversely affected
by the presence of clouds. In such scenarios, microwave data is more preferable due to the potential of penetrating through the clouds.
Recently launched (26 September 2016) scatterometer satellite (SCATSAT-1) data by the Indian Space Research Organization (ISRO)
has the potential of providing all-weather, day-night monitoring and daily data-delivery services at the global level. Along with the
numerous advantages, the Ku-band (13.535 GHz) based SCATSAT-1 cannot provide sufficient information as provided by the
multispectral optical sensors. Therefore, in the present work, the microwave-based SCATSAT-1 and optical-based MODIS (moderate
resolution imaging spectroradiometer) have been fused using the nearest-neighbour approach to examine its effects in cloud removal
and its applications in classification. The study has been performed over Himachal Pradesh, India. This study has also discussed the
impact of different classifiers such as artificial neural network (ANN), spectral angle mapper (SAM), support vector machine (SVM),
and random forest (RF), on the fusion of SCATSAT-1 (including backscattered coefficients, i.e. sigma-nought and gamma-nought at
HH and VV polarizations) and MODIS dataset. Experimental results have confirmed that the accuracy of implemented classified maps
significantly increases with the fusion of both datasets as compared to the individual implementation of SCATSAT-1- and MODISclassified maps. From quantitative analysis, the RF classifier performs better as compared to other classifiers, i.e. ANN, SAM, and
SVM on the fused dataset. This study has many applications in the near real-time monitoring of snow/ice, agriculture activities, and
hydrological studies.
