A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering in Water Engineering and Management

Thesis (M. Eng.) - Asian Institute of Technology, 2022

Fine-resolution satellite images provide useful information in Precision Agriculture (PA). The poor quality of satellite images during lousy weather days limits their field or farm-scale application, while unmanned aerial vehicle (UAV) images have higher resolution than satellite images. It can be used in crop monitoring to optimize plantation management and assess the optimum planting density more accurately. Moreover, the UAVs can perform scheduled flights to survey vegetation areas with accurate and updated maps to detect and deduce the root causes of detected instances of stress in the areas of plant growth. However, the resolution of images for the most widely used satellite images, such as Sentinel-2 (10 m by 10 m), includes vegetation canopies with varying degrees of heterogeneity within single pixels. In particular, the structured nature of rice fields comprising predominantly bare soil and weeds poses challenges when estimating the rice-level biomass and vigor from the coarse-resolution satellite images that present signals averaged over rice canopies and non-canopies (e.g., soil). The main objective of this study is to monitor and analyze crop growth, chlorophyll content, and nitrogen content in the paddy crop and identify critical zones for irrigation management using UAV-based images for informed decision-making at plot scale farming in Thailand. This study examined the relationship between the normalized difference vegetation index (NDVI), the triangular greenness index (TGI), and the green normalized difference vegetation index (GNDVI) utilizing extremely high resolution images captured by a UAV equipped with a multi-spectral imaging filter. The aggregated 2-cm-scale NDVI, TGI, and GNDVI values were compared with Sentinel-2 (10-m) products collected from a rice plot in Nakhon Pathom Province, Thailand, within 16 weeks. Results showed that when upscaling the high-resolution NDVI, TGI, and GNDVI values (by aggregating multi-spectral input bands for NDVI, TGI, and GNDVI) to coarser resolutions in a highly structured paddy field, all VIs correlated in the middle stage of crop growth. Also, it was found that the VIs from the UAV were lower than those from the Sentinel-2, mainly because the UAV had a higher resolution.