Time-space-angle scale effects and incorporation patterns in estimating rice LAI and leaf chlorophyll content by UAV multispectral remote sensing

Abstract

Estimation of leaf area index (LAI) and leaf chlorophyll content (LCC) of rice is of great scientific and practical value for precision agriculture and ecological research, and unmanned aerial vehicle (UAV) remote sensing technology offers an effective monitoring resource. Aiming at the current problems of time–space-angle scale effects and unclear fusion patterns, intensive UAV observations in the field were designed in this paper to acquire different time–space-angle multispectral data (observation altitude range of 50–250 m, time range of 9:00–16:00 local time, and the angles including east, west, south, north, and vertical perspectives) for rice in 12 periods. Through analyzing the effects of canopy normalized difference red edge index (NDRE) variations and time–space-angle scales on the accuracy of LAI and LCC estimation during typical rice periods, it was determined that the utilization of NDRE derived from strong sunlights (backscattering direction) is more conducive to the estimation of rice LAI, whereas NDRE from mild sunlights is more appropriate for the rice LCC estimation. Multi-period rice LAI and LCC were estimated using a deep regression model with multihead attention mechanisms (MHAR) through position embedding and understanding of global and local information. The results demonstrated that the highest model accuracy was achieved by the variable inputs of optimal strategy (the coefficients of determination (R²) = 0.89, the root mean square error (RMSE) = 1.29, relative RMSE (RRMSE) = 16.78 % for LAI estimation and R² = 0.85, RMSE = 2.06, RRMSE = 5.46 % for LCC estimation), significantly higher than shallow machine learning and deep neural network (DNN) models. Furthermore, the addition of other vegetation indices inputs did not substantially improve the model accuracy. The proposed time–space-angle fusion model provides valuable insights for UAV remote sensing crop monitoring.