Assessment of PlanetScope Spectral Data for Estimation of Peanut Leaf Area Index Using Machine Learning and...

Area of Science:

• Agricultural remote sensing
• Crop physiology
• Precision agriculture

Background:

• Leaf Area Index (LAI) is crucial for assessing crop growth and is vital for agricultural research and precision farming.
• PlanetScope imagery offers high revisit frequency and consistent, high-resolution multispectral data suitable for crop monitoring.

Purpose of the Study:

• To develop and compare regression models for estimating peanut LAI using PlanetScope spectral bands and vegetation indices (VIs).
• To evaluate the performance of Random Forest (RF), eXtreme Gradient Boosting (XGBoost), and Partial Least Squares Regression (PLSR) algorithms for LAI estimation.

Main Methods:

• Developed and compared RF, XGBoost, and PLSR models to estimate peanut LAI.
• Utilized PlanetScope spectral bands and thirteen vegetation indices (VIs) as predictor variables.
• Evaluated individual VIs and combined top-performing VIs for model calibration and validation.

Main Results:

• Vegetation indices (VIs) showed strong correlations with peanut LAI, with TSAVI and RTVIcore being top individual predictors.
• RF models incorporating the top six VIs achieved the highest estimation accuracy (R² = 0.844, RMSE = 0.858 m²/m²).
• PlanetScope VIs outperformed spectral bands alone, and combining bands with VIs decreased accuracy, highlighting the importance of variable selection.

Conclusions:

• Random Forest (RF) models integrating selected vegetation indices (VIs) provide a highly accurate and efficient method for estimating peanut Leaf Area Index (LAI).
• The findings support the use of RF for robust crop monitoring, reducing the need for multiple complex models in peanut cultivation.