Retrieving leaf area index in wheat fields using unmanned aerial vehicle (UAV)-based LiDAR and hyperspectral imagery

Leaf Area Index (LAI) is a dimensionless measure representing the total leaf area per unit ground area. As such, LAI is a key parameter in crop models, as it directly influences the photosynthetic activity of crops, affecting their growth, development, and yield predictions. It also reflects the canopy structure, which plays an essential role in how the plant responds to environmental stresses. In this study, we used UAV-based light detection and ranging (LiDAR) data and hyperspectral imagery (HSI) as two modalities to predict LAI in a total of 60 plots within 10 wheat fields of various cultivars in the dryland areas of Israel. Field LAI was assessed via two methods – destructive (Li3100C, Licor, Nebraska, USA) and optical (Li2200C, Licor, Nebraska, USA). The LAI in the wheat fields ranged from 0.25 m² m^–2 to 7.7 m² m^–2 (average LAI over the dataset was 1.5 m² m^–2). To predict LAI, we used LiDAR-derived metrics such as height-related metrics (height percentiles and bi-centiles, canopy relief ratio (CRR), max, average, and mode height), and 3-D variables (3-D profile index (3DPI), 3-D voxel index (3DVI), and convex hull volumes), as well as spectral vegetation indices, in five machine learning (ML) algorithms – simple linear regression (SLR), partial least squares regression (PLSR), support vector machine (SVM), Random Forest (RF), and Extreme Gradient Boosting (XGBoost). Single metric SLR resulted in R² ranging from 0.32 to 0.55. More complex algorithms showed that the LiDAR-derived metrics were useful for estimating LAI at the plot level with a higher R² > 0.81 and an RMSE of less than 0.16 m² m^–2 (c. 10%) for the ML algorithms. The 3-D variables were shown to be the most important and robust variables in the ML models for predicting LAI at the plot level, with some height-related variables showing great potential as well. This study is a unique first-step effort to evaluate UAV-LiDAR sensors in collecting high-quality, non-destructive, repeatable measurements of LAI. Such remote sensing information could be highly useful to calibrate and evaluate crop models while resolving the upscaling limitation from leaf to canopy.