Aboveground Biomass Estimation of Rice Crops (Oryza sativa L.) Utilizing Parameters Derived From UAV-Based LiDAR And Multispectral Satellite Sensors

Abstract. This study investigates efficient, non-destructive approaches for estimating rice aboveground biomass (AGB), a key yield indicator. It integrates Unmanned Aerial Vehicle - based Light Detection and Ranging (UAV-LiDAR) sensor for structural data and multispectral satellite imagery for spectral data to develop individual and fused models aimed at improving AGB estimation accuracy. Data were collected across three rice growth stages during one planting season for National Seed Industry Council (NSIC) Rc 222 and NSIC Rc 160 rice cultivars, using UAV-LiDAR, PlanetScope imagery, and field-based AGB measurements, wherein 30 samples were used for analysis. Multiple linear regression was used to model fused spectral and structural parameters for each variety. Results showed model performance depends on rice variety. Through Leave-One-Out Cross-Validation (LOOCV) and the corrected Akaike Information Criterion (AICc), the spectral-only model for NSIC Rc 160 using Green Normalized Difference Vegetation Index (GNDVI) performed best (R²=0.62, RMSE=5.16, rRMSE=1.85%, AICc=187.10). Structural data did not improve the model. For NSIC Rc 222, the fused model combining GNDVI, Normalized Difference Yellowness Index (NDYI), and canopy height achieved the highest accuracy (R² = 0.82, RMSE=10.40, rRMSE=5.86%, AICc=165.60), indicating that combining spectral and structural data enhances predictions. Due to the small sample size, LOOCV was used, but larger datasets are needed to explore advanced machine learning methods. These findings support modeling approaches per rice variety and highlight its potential for precision agriculture applications in rice biomass estimation.