Mowing detection from combined Sentinel-1, Sentinel-2, and Landsat 8 time series on fallow cropland with transfer learning

Abstract

With the monitoring of management practices in agriculture we can assess the influence on the environment and evaluate the implementation of policies and guidelines. The timing of mowing or mulching events on fallow cropland is a critical factor in assessing the impact on biodiversity and a core environmental measure of the EU's Common Agricultural Policy (CAP). Dense time series of remote sensing data have already proven to be valuable input for capturing management practices on grassland. Here we present a transfer learning approach using a 1D convolutional neural network trained on grassland management data to predict management practices on fallow cropland. Our results demonstrate the potential of the approach to provide valuable insights for future CAP decision-making.