Crop Type Mapping Case Study in the Lower Mekong Basin based on Transfer Learning Techniques using GEDI and Multispectral Earth Observation Data

Crop type mapping is essential for improving the agricultural monitoring practices mitigating food security issues due to climate change. High-accuracy crop type maps are difficult to produce in locations where in-situ data is sparse. Reliable maps are commonly made using methods and products confined to local specifications and where ground labels are available. Expanding on the work by Tommaso et al., 2021, this research conducts a case study by applying the same methodology in the Lower Mekong using NASA's Global Ecosystem Dynamics Investigation (GEDI) spaceborne lidar to distinguish maize and other tall and short crops with high accuracy and improved transferability across regions compared to optical features alone. This study then compares the method's transferability for detecting other crop types, in data sparse regions throughout the Lower Mekong between 2019-2021, assessing whether additional resources can improve mapping accuracies without, or limited, regional crop type labels. These findings will augment NASA-USAID SERVIR Program's existing services estimating rice yield, and can inform how new satellite observations complement sparse in-situ data for food security monitoring systems used by regional decision makers.