Precise identification of maize in the North China Plain based on Sentinel-1A SAR time series data

Abstract

ABSTRACT Playing a vital role in the national economy and food security of many countries, maize should be identified precisely to accurately determine planting area and yield times. Analysis of multi-temporal backscatter characteristics based on Synthetic Aperture Radar (SAR) data provides an effective approach to spatially map maize on a regional scale. Accordingly, we acquired nine European Space Agency (ESA) Sentinel-1A C-Band SAR dual-polarimetric (VV and VH) radar time series data at a 20 m spatial resolution for a segment of the North China Plain in 2016. After observing variability in C-Band VV and VH backscattering coefficients and analysing microwave backscattering characteristics for different structures at different growth and development stages, as well as a comparison between different crops, we selected the optimum temporal phases to construct Sentinel-1A SAR time series data sequences for classification. To assess the performance of different polarization modes in identifying maize, both single- and dual-polarimetric data were used. Results showed that a high-recognition accuracy could be achieved for maize using SAR time series data. Specifically, long-sequenced dual-polarimetric SAR time series data performed better with an overall accuracy of 93.57% and a kappa coefficient of 0.92 compared to dual-polarimetric SAR data from July (45.13% and 0.34, respectively). Although an increment increase in the number of images used improved overall accuracy, more than five could result in an increase in errors. Single-polarimetric data from a single temporal phase yielded an accuracy of only 31.58%. Additionally, classification accuracy obtained by applying dual-polarization could be improved by 13% compared to single-polarimetric SAR data from a single temporal phase time acquisition alone. However, the differences among the single-polarization, dual-polarization, and multi-polarization results were relatively moderate when sufficient temporal phase data were applied.