Time-weighted dynamic time warping analysis for mapping interannual cropping practices changes in large-scale agro-industrial farms in Brazilian Cerrado

Abstract

Methods for crop phenology detection using time series analysis have provided accurate information for large agricultural areas in shorter processing times, which can be useful for agronomic management and supply chain monitoring. Given the crop dynamics in the Brazilian Cerrado, with alternating crop type plantings, crop successions, and crop rotations, as well as climate and crop practices variation between harvest periods, these methods can be useful for detecting subtle land use and land cover changes at farm and crop field scales, improving thematic classifications and the near real-time crop monitoring. In this study, the Time-Weighted Dynamic Time Warping method was applied to recognize patterns in Moderate Resolution Imaging Spectroradiometer (MODIS) time series for land use and cover classification, identifying crop successions and rotations at crop field level in a large-scale agro-industrial agglomerate of farms located at Brazilian Cerrado. We detected and analyzed temporal cropping patterns in training samples to classify the MODIS time series and images, using a robust ground truth data set for validation. The method distinguished Cotton-fallow, Soybean-cotton, Soybean-maize, and Soybean-millet cropping patterns with an overall accuracy above 85% for all evaluated harvest periods. Seasonal variations in the crop fields, caused by interannual succession and rotation, were detected. The method demonstrated the benefit of creating a spatial vector data set for supporting decision-making in several crop management contexts, improving crop and supply chain monitoring.