The PROSPECT model in high-throughput phenotyping for peanut leaf parameter estimation: Comparative performance of hyperspectral inversion models

Abstract

Accurate estimation of leaf biochemical parameters is crucial for understanding crop physiology and monitoring nutritional status. Remote sensing algorithms perform well on limited germplasm, but the transferability to high-throughput phenotyping with diverse genotypes remains unclear. This study estimated leaf chlorophyll content (Cab), equivalent water thickness (Cw), and dry matter content (Cm) using the single vegetation index (SVI), random forest (RF), and the PROSPECT model to evaluate the performance and transferability of these models under diverse peanut germplasm conditions. Results showed that Transformed Chlorophyll Absorption in Reflectance Index (TCARI), Water Index (WI), and Modified Simple Ratio (mSR) were strongly correlated with Cab, Cw, and Cm, respectively, highlighting their importance in the inversion models. Comparative analysis revealed that the RF model achieved the highest accuracy for Cab (R² = 0.77, RMSE = 8.14 µg cm⁻²), Cw (R² = 0.67, RMSE = 1.1 × 10⁻³ g cm⁻²), and Cm (R² = 0.50, RMSE = 6.2 × 10⁻⁴ g cm⁻²), followed by the PROSPECT model, with R² and RMSE of 0.76 and 8.21 µg cm⁻² for Cab, 0.61 and 1.2 × 10⁻³ g cm⁻² for Cw, and 0.38 and 7.7 × 10⁻⁴ g cm⁻² for Cm, respectively. However, the PROSPECT model was most effective in Cab inversion across diverse germplasm resources (R² = 0.58, RMSE = 7.68 µg cm⁻²), demonstrating its superior transferability and stability. These results underscore its value in high-throughput phenotyping and improving the accuracy and generalizability of crop biochemical parameter estimation.