Correlation of yield and vegetation indices from unmanned aerial vehicle multispectral imagery in Thailand rice production systems

Abstract

Unmanned aerial vehicles (UAVs) equipped with cameras are used for collecting vegetation indices (VIs) for health monitoring of field crops such as rice (Oryza sativa). This study evaluated the relationship of VIs derived from multispectral UAV images at different buffer zones around the sampling points with rice biomass and grain yield from 20 nonirrigated and irrigated fields in Sakhon Nakhon, Thailand, in 2021. Varying nitrogen (N) rates (21.8–98.7 kg ha−1) were applied in splits, at 14 days after transplanting and panicle initiation (PI) stage. One week after PI, multispectral images were captured by a DJI Phantom 4 Multispectral UAV before taking biomass samples at four 1-m2 sampling points for the three plots in each field. At harvest, whole plant samples were collected from nearby these sampling points for grain yield estimation. For each sampling point at buffer zones 5, 10, and 20 m, the average of four VIs (normalized difference vegetation index [NDVI], green NDVI [GNDVI], normalized difference red-edge [NDRE], and optimized soil adjusted vegetation index [OSAVI]) was computed. Correlation analysis showed NDRE had the highest correlation with grain yield in nonirrigated systems (r = 0.575–0.613) and pooled data (r = 0.523–0.539). NDVI moderately correlated with biomass (r = 0.224–0.233). Images within the 10-m buffer zone produced NDRE values most strongly linked to yield, both unadjusted and normalized by planting to sensing days and growing degree days. There is a potential to use NDRE as predictor of rice biomass yield in rice systems.