Accurate inversion of chlorophyll content based on PROSPECT-LSROGF-BAS-BP method

Abstract

Accurate measurement of chlorophyll content in plant leaves is crucial for evaluating plant health. Leaf radiation transfer models are commonly used to estimate chlorophyll content from remote sensing data. However, current methods often show limited accuracy in certain scenarios. This study addresses these challenges by developing a more precise method for chlorophyll content retrieval. First, the PROSPECT model, which does not fully account for optical reflection on leaf surfaces, results in lower spectral simulation accuracy. To overcome this limitation, a surface geometric feature factor (σ) is introduced, leading to the PROSPECT-LSROGF model. This enhanced model incorporates the optical reflection characteristics of the leaf surface, expands the range of light source incident angles, and more accurately describes radiative transfer within the leaf. As a result, the PROSPECT-LSROGF model shows superior spectral simulation accuracy to the traditional PROSPECT and PIOSL models. Next, to improve the retrieval accuracy of traditional BP neural networks for chlorophyll content, the Beetle Antennae Search (BAS) algorithm is used to optimize the weights and thresholds of the BP neural network, forming the BAS-BP model. By combining the PROSPECT-LSROGF model with the BAS-BP network, the PROSPECT-LSROGF-BAS-BP model is developed for accurate chlorophyll content retrieval. The performance of this model is compared with that of the gradient boosting machine retrieval and the PROSPECT-BAS-BP model. Validation is conducted using the LOPEX93, CABO, and ANGERS datasets. The PROSPECT-LSROGF-BAS-BP model achieves root mean square errors (RMSEs) of 4.186, 4.258, and 3.894 g/cm2, with determination coefficients (R2) of 0.876, 0.862, and 0.903, respectively—outperforming the other methods in terms of accuracy. These results demonstrate that the proposed method significantly improves the model’s ability to accurately estimate chlorophyll content from spectral data.