An optimum datasets analysis for monitoring crops using remotely sensed Sentinel-1A SAR data

Abstract

ABSTRACT To effectively monitor crops, it is necessary to select extremely redundant satellite images and to know the number of acquisitions required for a specific period to analyse cropping patterns, thereby reducing analysis time. In this paper, we have examined an empirical analysis for the optimum dataset (OptD) selection required to monitor the crops. Sentinel-1 dual-polarized SAR datasets were used in this study to illustrate the effectiveness of optimum datasets required for the considered crops (ginger, tobacco, rice, cabbage, and pumpkin). In this work, at first, the entropy and alpha bands were treated as cluster centres for crop decomposition and its scattering mechanism using the cluster-based K-means unsupervised classification technique. The clusters are plotted on the H-α plane to get the H-α plot of dual-polarization SAR data for target decomposition. To understand the dominance of scattering type with crop growth stage, the obtained scattering distribution from the H-alpha plot is scaled to a percentage analysis. Based on qualitative observations of the percent scattering distribution of crop pixels over a h-alpha plot and backscattering coefficient behaviour at different crop growth stages, an empirical approach has been used to select dataset reduction. It has been suggested that the combination of successive repeated data with similar scattering analysis of combined h-alpha plots and backscattering analysis is the best reduced dataset selection for effective crop monitoring. From the analysis, the optimum dataset required for monitoring Ginger (from the flourishing stage), Tobacco, Paddy, Cabbage, and Pumpkin has been identified, and found that the tobacco crop requires fewer datasets, whereas the rice crop requires a greater number of datasets for monitoring. Despite the challenges associated with, p-bias for the crops was achieved at good levels, given that, lowering the datasets to obtain the optimal number without significantly reducing the accuracy of the data.