Estimating leaf phosphorus concentration in rice by combining vegetation indices, texture features, and water indices from UAV multispectral imagery

Abstract

Leaf phosphorus (P) concentration is a key factor that reflects the growth of rice (Oryza sativa), affecting both the quality and productivity of the crop. The estimation of leaf P concentration using unmanned aerial vehicle (UAV) remote sensing plays a pivotal role in fertilization management, monitoring rice growth, and advancing precision agriculture strategies. This study aimed to integrate vegetation indices (VIs), texture features (TFs) indices, and water indices (WIs) obtained from UAV multispectral images to estimate leaf P concentration in rice using the multi-criteria evaluation (MCE) model with analytical hierarchy process–based weights. The MCE method was employed to integrate the 16 VIs, eight TFs, and two WIs with four scenarios (S1, S2, S3, and S4) to evaluate their contributions to estimating the rice leaf P concentration. The S1 integrates the normalized difference vegetation index (NDVI), the modified chlorophyll absorption in reflectance index (MCARI), and the mean (MEA). The S2 extends S1 by incorporating the normalized difference water index (NDWI), while S3 combines the indices from S1 with NIR shoulder region index (NSRI). Finally, S4 integrates NDVI, MCARI, MEA, NDWI, and NSRI. The S4, which integrates all VIs, TFs, and WIs, provides the highest accuracy in estimating leaf P concentration with root mean square error values of 0.035. The research findings indicate that leaf P concentration differs between the two rice varieties, TBR225 and J02. The J02 variety exhibits a higher leaf P concentration than the TBR225 variety, as it is more efficient in P synthesis. The results of this study provide an effective foundation for developing solutions in rice nutrition management, with a focus on advancing precision agriculture.