COMPARISON OF CROP DISCRIMINATION USING AVIRIS-NG AND LISS-IV DATA OVER HETEROGENEOUS AGRICULTURAL PATCHES

摘要

Crop mapping and discrimination provide an important basis for many agricultural applications such as acreage, biomass, yield, crop rotation and soil productivity. Remote sensing data, methods and approaches provide the best options for large area agricultural cropland characterization for precision agricultural management practices by accurately mapping of crop type and yield indicators. Traditional multispectral broadband sensor data have known limitations of sensor saturation and absence of specific narrow bands to target and highlight specific biophysical and biochemical characteristics according to crop type. These factors lead to significant uncertainties in the discrimination of crop type. Recent advances in hyperspectral remote sensing technology provide the opportunity to measure the response of different crop type in terms of morphological and physiological characteristics. The specific narrow bands have a capability to perform crop discrimination over homogeneous and heterogeneous agricultural areas. The continuous band spectrum from imaging spectroscopy have opened up new avenues in the field of classification. In this study, crop discrimination has been carried out using principal component analysis and supervised classification techniques such as maximum likelihood classification (MLC) and spectral angle mapper (SAM) algorithms. In this study, AVIRIS-NG airborne hyperspectral data acquired on Maddur. Karnataka and equivalent multispectral LISS-IV data convolved through three broadband regions (Green: 0.52-0.59 run, red: 0.62-0.68 run, near-infrared: 0.77-0.86 nm) using spectral response function of Resourcesat-2 (RS-2) LISS-IV, were used over mixed and heterogeneous agricultural area of Maddur, Karnataka in Berambadi watershed located in Kabini river basin. The dominant soil types were red and black soils. Data dimensional reduction has been carried out using principal component analysis. In situ crop information were used to perform SAM and MLC-based classification. Classification accuracy was computed using confusion metrics. SAM classification showed classification accuracy of the order of 77.7 % and 42.8% with Kappa coefficient of 0.75 and 0.34 for AVIRIS-NG and LISS-IV equivalent, data, respectively. The MLC-based classification showed accuracy of 94.3% and 55.6% and Kappa coefficient of 0.93 and 0.46 for AVIRIS-NG and LISS-IV data. It can be concluded that imaging hyperspectral narrowband data has the potential to discriminate crops in a mixed and heterogeneous crop cluster with higher accuracy as compared to equivalent resolution multi-spectral broadband data.