Wheat leaf area index retrieval from drone-derived hyperspectral and LiDAR imagery using machine learning algorithms

Leaf Area Index (LAI) is a key parameter that reflects canopy structure and influences photosynthetic activity. Traditional remote sensing methods using spectral indices usually struggle with saturation at LAI > 3.0 m² m^–2 in crop fields. Light detection and ranging (LiDAR) systems offer a solution by capturing detailed canopy structures. This study used drone-based LiDAR and hyperspectral imagery to predict LAI across 60 plots in five wheat fields in Israel. Field LAI, assessed using a handheld optical sensor, ranged from 0.25 to 7.7 m² m^–2. LiDAR-derived metrics, including height, gap fraction, and canopy volume features, were combined with spectral indices for LAI prediction. These metrics were used in simple linear regression (SLR) and five machine learning (ML) models: artificial neural network (ANN), random forest, ridge regression, relevance vector machine, and extreme gradient boosting. Shapley's additive explanations identified key predictive features. Results show that ML models significantly improved prediction performance (R² = 0.59–0.90) compared to single metric SLR models (R² = 0.09–0.67). Combined LiDAR-spectral models outperformed spectral- and LiDAR-only models. ANN achieved the best results, with a mean R² of 0.90, normalized RMSE of 6 %, and residual prediction deviation (RPD) score of 3.34, accurately predicting LAI up to 5.5 m² m^–2. LiDAR alone or in combination with spectral metrics improved LAI predictions. While some spectral metrics ranked high, LiDAR-derived metrics, particularly canopy volume-related, consistently emerged among the most important features, with gap fraction and height metrics also contributing to the models. This study demonstrates the efficacy of drone-based LiDAR for non-destructively predicting LAI in wheat fields, offering a valuable tool for crop model calibration and evaluation and addressing the challenge of scaling from leaf to canopy.