Deep learning solutions for very high resolution mapping of contour levee rice production systems

Authors: DALE, DAKOTA - University Of North Texas; LIANG, LU - University Of North Texas; ZHONG, LIHENG - Descartes Labs, Inc; Reba, Michele; RUNKLE, BENJAMIN - University Of Arkansas

Submitted to: Computers and Electronics in Agriculture
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 4/24/2023
Publication Date: 6/17/2023
Citation: Dale, D., Liang, L., Zhong, L., Reba, M.L., Runkle, B. 2023. Deep learning solutions for very high resolution mapping of contour levee rice production systems. Computers and Electronics in Agriculture. 211. Article 107954. https://doi.org/10.1016/j.compag.2023.107954.
DOI: https://doi.org/10.1016/j.compag.2023.107954

Interpretive Summary: The identification and mapping of irrigation systems is an important task for managing and predicting the water, carbon, energy, and productivity implications of agricultural practices. This work is particularly important in rice production due to this crop’s global importance and environmental impact. In our submission we develop and test a novel deep learning system capable of identifying crop fields with contour style levee irrigation systems from an aerial imagery product. The model performs well even under the introduction of noise such as clouds, Gaussian Static or reductions in image resolution. The model thus demonstrates potential to function given different spatial extents or even using other imagery datasets. The work will enable a comprehensive analysis of landscape use patterns to drive models of agricultural productivity and sustainability. It should also offer a guide for approaching any type of deep learning pattern identification problem, no matter the discipline.

Technical Abstract: The construction of contour levees for rice irrigation represents a major landscape management practice with impacts on irrigation water use efficiency, crop management decisions, and food production. However, levee distributions are traditionally reliant on local field surveys and remote sensing approaches are complicated by irregular spacing, shape, and landscape variability within the field itself. In this paper the authors develop a deep learning approach capable of identifying rice fields with contour levee irrigation practices from open source aerial imagery. To generate a levee-identification scheme, a hybrid ResNet/Unet model is built from the commonly known Residual Network (ResNet) architecture for multi-layer deep learning strategies. The model takes a 320 ' 320 RGB aerial landscape image from the US National Agricultural Imagery Program as input along with label data to then generate a probability map of the distribution of farm fields that use contour levees. In performing this task, the model generates a 0.959 receiver operating characteristic curve score. The model continues to perform well under the introduction of clouds, data augmentation, or reductions in spatial resolution. Throughout these tests, the model performed within 0.03 of its original score, except for when the image quality was reduced to one fifth of its original resolution wherein the model score dropped to 0.915. Via these tests the model demonstrates potential to function well given different spatial extents or potential satellite remote sensing with moderate (10m) resolutions.