Detecting Waterlogging in Agricultural Fields in Denmark using High-Resolution PlanetScope Time Series

Waterlogging in agricultural fields is the condition of temporally inundated areas driven by extreme rainfall, rising groundwater or poor drainage, and has been identified as a major issue by Danish farmers. During the inundation period, plants are deprived of oxygen which negatively affects the root development and leads to decreased yields and grain quality. Additionally, these waterlogged areas are a large source of greenhouse gas (GHG) emissions. The issue is expected to exacerbate under current climate projections through wetter winters and rising groundwater levels in Denmark. Hence, an increased understanding of the spatio-temporal dynamics of waterlogging is required to future-proof the management strategies. The research goals are three-fold: (1) to optimise the detection of waterlogging, (2) to reveal inter- and intra-annual patters across Denmark and (3) to investigate the drivers of waterlogging such as climate, topography and bio-physical conditions. We aim to detect waterlogged areas through a deep learning semantic segmentation approach utilising multi-temporal PlanetScope imagery and nation-wide high resolution elevation data. This approach requires a manually delineated reference dataset to train, validate and test the model which needs to be well-balanced spatially, e.g. covering various soil types, and temporally, e.g. including various illumination conditions. Additionally, we will experiment with various model architectures, backbones and covariate combinations to optimise the segmentation performance. Initial tests using a UNET architecture and building upon a published reference dataset by Elberling et al. (2023), show promising results and lay the foundation for the upcoming model development and extension of the existing reference data.

 

^{Elberling, B. B., Kovacs, G. M., Hansen, H. F. E., Fensholt, R., Ambus, P., Tong, X., ... & Oehmcke, S. (2023). High nitrous oxide emissions from temporary flooded depressions within croplands. Communications Earth & Environment, 4(1), 463.}