Kharif crop characterization using combination of SAR and MSI Optical Sentinel Satellite datasets

Abstract

In the present study, the differences in the kharif crop reflectance at varied wavelength regions and temporal SAR backscatter (at VV and VH polarizations) during different crop stages were analyzed to classify crop types in parts of Ranchi district, East India using random forest classifier. The spectral signature of crops was generated during various growth stages using temporal Sentinel-2 MSI (optical) satellite images. The temporal backscatter profile that depends on the geometric and di-electric properties of crops were studied using Sentinel-1 SAR data. The spectral profile exhibited distinctive reflectance at the NIR (0.842 µm) and SWIR (1.610 µm) wavelength regions for paddy (Oryza sativa; ~0.25 at NIR, ~0.27 at SWIR), maize (Zea mays; ~0.24 at NIR, ~0.29 at SWIR) and finger millet (Eleusine coracana, ~0.26 NIR, ~0.31 at SWIR) during pre-sowing season (mid-June). Similar variations in crop’s reflectance at their different growth stages (vegetative to harvesting) were observed at various wavelength ranges. Further, the variations in the backscatter coefficient of different crops were observed at various growth stages depending upon the differences in sowing–harvesting periods, field conditions, geometry, and water presence in the crop field, etc. The Sentinel-1 SAR based study indicated difference in the backscatter of crops (i.e., ~−18.5 dB (VH) and ~−10 dB (VV) for paddy, ~−14 dB (VH) and ~−7.5 dB (VV) for maize, ~−14.5 dB and ~−8 dB (VV) for finger millet) during late-July (transplantation for paddy; early vegetative for maize and finger millet). These variations in the reflectance and backscatter values during various stages were used to deduce the best combination of the optical and SAR layers in order to classify each crop precisely. The GLCM texture analysis was performed on SAR for better classification of crop fields with higher accuracies. The SAR-MSI based kharif crop assessment (2017) indicated that the total cropped area under paddy, maize and finger millet was 24,544.55, 1468.28 and 632.48 ha, respectively. The result was validated with ground observations, which indicated an overall accuracy of 83.87% and kappa coefficient of 0.78. The high temporal, spatial spectral agility of Sentinel satellite are highly suitable for kharif crop monitoring. The study signifies the role of combined SAR–MSI technology for accurate mapping and monitoring of kharif crops.