In agricultural and remote sensing research, the precise estimation of Wheat's Leaf Area Index (LAI) through the utilization of unmanned aerial vehicle (UAV)-derived multispectral data is crucial for effectively monitoring crop health and growth.
Research flow map. Image Credit: TranSpread
Traditionally, the measurement of LAI has been accurate but labor-intensive. Recent progress has seen the emergence of hybrid methods that combine radiative transfer models with machine learning, displaying promise due to their efficiency and versatility.
However, these methods encounter challenges, particularly in diverse soil backgrounds, where scalable soil-specific models are lacking.
Current research is centered on crafting a “background-resistant” model to achieve stable and accurate LAI estimation across various soil types and environmental conditions, particularly beneficial in regions with variable soil characteristics and low LAI, such as dryland areas.
The study was published in the journal Plant Phenomics.
The aim of this research was to create a generic machine learning-based model for predicting wheat Leaf Area Index (LAI) across diverse soil backgrounds throughout the entire growth season, surpassing previous soil-specific models. The model’s simulation performance was initially tested on independent synthetic data.
Random Forest Regression (RFR) models, trained on synthetic data, exhibited varying performances based on soil reflectance similarity. The baseline model achieved an R² of 0.8 on similar soil reflectance but dropped to 0.2 on dissimilar soils.
Expanding the reflectance domain of the training soil background enhanced the model's robustness, but improving canopy-spectral inputs proved more effective for stable LAI prediction across soil backgrounds.
The RFR models underwent testing using synthetic and augmented data across various growth stages in experiments. The enhancement in predicting LAI was notably higher when focusing on enhancing canopy-spectral inputs rather than merely broadening the reflectance domain of the training soil background.
Among the models, the defaultMulti2.VIc3 model, which employed an expanded reflectance domain and enhanced canopy-spectral indicators, stood out for its stability across different soils and required fewer input variables, thus warranting further evaluation.
The model demonstrated good estimation accuracy for various soil backgrounds but tended to overestimate LAI for values between 2 and 5 and underestimate LAI over 5.
Further evaluation at different growth stages throughout the growing season revealed substantial improvement in prediction accuracy, especially at early and late stages. It reliably captured the seasonal LAI dynamics under different treatments concerning genotypes, planting densities, and water-nitrogen management.
The study found that employing simulation data enables the creation of a robust background-resistant model. This model delivers consistent and precise GAI prediction from individual UAV-based multispectral images throughout a wheat-growing season, even in diverse field conditions. This advancement in LAI prediction, devoid of ground calibration, signifies a valuable tool for agricultural monitoring and management.
Journal Reference:
Chen, Q., et al. (2023) A Generic Model to Estimate Wheat LAI over Growing Season Regardless of the Soil-Type Background. Plant Phenomics. doi.org/10.34133/plantphenomics.0055.
Source: https://www.transpread.com/