Monitoring Changes In Agricultural Field Boundaries Using Spatiotemporal Remote Sensing Data

In the domain of precision agriculture, land-use planning, and resource management, the precise delineation of field boundaries is pivotal for informed decision-making. The dynamic nature of agricultural landscapes, particularly in smallholder farming, introduces seasonal changes that pose challenges to accurately identify and update field boundaries. The conventional approach of relying on high-resolution imagery for this purpose proves to be economically impractical on a seasonal basis. We propose a framework that utilizes a spatiotemporal series of medium-resolution public imagery (e.g., Sentinel-2) in conjunction with an outdated high-resolution image as a reference for super-resolution reconstruction. The developed methodology incorporates super-resolution techniques to enhance the spatial resolution while simultaneously performing semantic segmentation at the higher resolution. We evaluate the proposed model's performance in predicting seasonal field boundaries at a pan-India level. The validity of these findings is established through assessment by a team of human annotators.

Our approach aims to offer a scalable spatiotemporal solution for accurate field boundary identification at a national level by combining information from different satellites at different resolutions.