[M-AG32] Satellite Land Physical Processes Monitoring at Medium/High/Very High Resolution

[MAG32-P11] Commercial high-spatial resolution satellite imagery for crop field extraction in Asian smallholder regions

Lin Yan¹, *David Roy² (1.Center for Global Change and Earth Observations, Michigan State University, East Lansing, MI 48824, USA, 2.Center for Global Change and Earth Observations & Depratment of Geography, Michigan State University, East Lansing, MI 48824, USA)

Keywords:agricultural fields, object based , field extraction , subsistence and smallholder agriculture

Delineated field boundaries, field sizes, and their change, are needed to understand agricultural activity, for land use planning, and local-to-national scale agricultural modelling. Automated extraction of field objects from satellite data is an active research topic. Field extraction is easier where the fields have straight, well-maintained boundaries, and the side lengths are an order of magnitude greater than the satellite pixel dimensions. These characteristics are associated typically with industrial mechanized farming. In regions with subsistence and smallholder agriculture, fields often cannot be distinguished in medium-resolution satellite imagery because they are small, irregularly-shaped, have narrow margins between adjacent fields, and may exhibit low spectral contrast with neighboring fields. The availability of commercial high spatial resolution (<10 m) satellite data provides new opportunities for field extraction. A new extraction algorithm based on our published Landsat-based Conterminous United States field extraction algorithm is presented. The algorithm implements an edge detection model that simulates human visual cortex edge perceptions, and is shown to (i) effectively capture fine-scale edge structure while being insensitive to interior field variations commonly exhibited in high resolution images, and (ii) is robust to irregular field shapes and obscure field boundaries that are commonly observed in smallholder fields. Crop field extraction results derived using seasonal WorldView (1-2 m) and PlanetScope (3-4 m) imagery over smallholder regions in the Central Plain of Thailand, the Jianghan Plain in mid-southern China, and also over small fields in Louisiana, USA are presented.