Project Database

This proposal aims to develop products based on satellite remote sensing imagery to resolve several sustainable agriculture issues in Taiwan, including national crop inventory automation as a foundation for long-term crop planting structure adjustment, rice yield prediction to better address agricultural losses, and early-season rice field mapping for a better-informed water resource management. From a remote sensing perspective, Taiwan’s agriculture possesses two unique challenges: small parcel sizes (<0>40%). To tackle these challenges, the central idea of this project is to jointly utilize synthetic aperture radar (SAR) and optical satellite imagery, in particular with SAR data of different polarization modes, radar frequencies, and spatio-temporal resolutions. Most existing studies only rely on the open-access, medium-resolution Sentinel-1 data on agricultural applications, but we argue that it is far less than enough in terms of the accuracy and precision needed. Instead of relying a single data source, we seek for both open and non-open SAR data, multi-frequency and multiple polarization modes, and aim for producing different SAR polarimetric descriptors that can be used in combination with optical indices. We will adopt different machine learning architectures to achieve the proposed products, with the goal of providing feature selection and uncertainty estimation during the product development. These products and side products will help provide more accurate and real-time information at a large scale even when there is limited manpower available, and hence the decision about crop types, yield losses and water requirements can be better adapted to the rapidly changing climate in the upcoming future.