A hybrid time-weighted dynamic time warping and TKNN classifier for paddy rice mapping using time-series Sentinel-1 SAR data

Abstract

Accurate rice extraction is essential for agricultural monitoring. However, traditional methods using Euclidean Distance (ED) or Dynamic Time Warping (DTW) are limited by noise sensitivity and ignoring temporal order. This paper proposes a new method combining Time-Weighted Dynamic Time Warping with Topology-based K-Nearest Neighbour (TWDTW-TKNN) for rice mapping using SAR time-series data. The TWDTW metric effectively fuses temporal dynamics, improving classification robustness. Experiments in Wuwei, Anhui, using 2020 SAR data, demonstrate that the proposed model achieves superior accuracy in noise suppression and temporal feature retention, outperforming conventional distance metrics. The method shows strong potential for time-series remote sensing classification.