Time-Weighted Dynamic Time Warping Classification Algorithm for Land Cover Mapping by Using SAR Imagery

Small to marginal landholdings with a diversity in cropping and management practices pose formidable challenges to implement crop classification algorithms. Satellite imagery captured in the optical bands is often contaminated with cloud cover and fails to detect the phenological as well as the structural changes happening during the crop growth period. This is particularly true with Indian agro-climatic settings, wherein the crop cycle largely overlaps with the monsoon season. This work is aimed at developing a novel crop classification algorithm that utilizes the temporal patterns of synthetic aperture radar (SAR) datasets from Sentinel-1 to spatially map the heterogeneous, fragmented croplands. Radar Vegetation Index is considered to develop the temporal crop patterns and correlate with un-classified time series of satellite imagery using time-weighted dynamic time wrapping (TWDTW) algorithm. Applicability of the proposed algorithm was tested in south India, to identify four crop varieties. Ground truth data was utilized to develop the temporal signatures of the training sets. It was observed that parcel-based classification using TWDTW algorithm yielded an overall accuracy of 63% (Kappa coefficient = 0.61). The results conclude that SAR-derived indices can effectively classify the croplands for use with yield and damage assessment modelling studies in agriculture.