Integration of a parameter combination discriminator improves the accuracy of chlorophyll inversion from spectral imaging of rice

Abstract

The PROSPECT model, widely employed for leaf radiation transfer analysis, relies heavily on input biochemical parameters to calculate spectral reflectance. This dependence often results in similar simulated spectra for different parameter combinations, which complicates the inversion of leaf chlorophyll content (Cab). To address this ill-posed problem, we enhanced the model's application by integrating a support vector machine (SVM)-based parameter combination discriminator with the Look-Up Table (LUT) constructed from the PROSPECT model. We marked samples in the LUT to reflect their closeness to measured parameters, facilitating the identification of reasonable versus unreasonable parameter combinations. The discriminator could effectively discriminate between reasonable and unreasonable parameter combinations, achieving accuracies of 0.894 and 0.888 in the training and test sets, respectively. The discriminator was then employed to refine the LUT, and an improved third-generation non-dominated ranking genetic algorithm (NSGA-III) was used to optimize the extreme learning machine. The inversion of rice Cab using the refined LUT and the NSGA-III demonstrated substantial improvements. The LUT was significantly improved after integration with the discriminator, yielding R2 and RMSE of 0.665 and 7.220 ​μg ​cm−2, respectively. The NSGA-III inversion, which utilized the “constraint method” with discriminator results as optimization objectives, achieved the best inversion accuracy, with R2 and RMSE values of 0.809 and 4.788, respectively. This study demonstrates that the effective use of a parameter discriminator can significantly enhance the accuracy of Cab inversion based on the PROSPECT model, offering a substantial advancement in addressing its inherent ill-posed challenges.