Hyperspectral-Based Retrieval of Clumping Index in Natural Secondary Forests Using TPOT Algorithm

The clumping index (CI) is a key parameter for quantifying foliage aggregation within vegetation canopies and is essential for accurately characterizing related physiological and ecological processes. Critically, even with identical leaf-level biochemical parameters, differences in canopy structure can yield distinct reflectance signatures, underscoring the role of CI in reducing uncertainties in retrieving vegetation biochemical properties. Traditional CI estimation often relies on coarse-resolution, multi-angular MODIS data, whose limited spectral and spatial resolution hinders accurate trait retrieval and reduces accuracy. In this study, a novel CI inversion framework that leverages the rich spectral information from satellite-based hyperspectral imagery ZY1F-02E data was developed to estimate CI timely and conveniently. CI was measured using TRAC with the LX+ method across 44 forest plots in Maoer Mountain, Heilongjiang Province. Three machine learning algorithms—Support Vector Machine (SVM), Random Forest (RF), and Tree-based Pipeline Optimization Tool (TPOT)—were compared for CI estimations. The optimal model was selected based on 5-fold cross-validation accuracy and uncertainty analysis. Results indicate that: (1) vegetation indices derived from second-derivative spectra (e.g., mRVI and EVI) exhibited stronger correlation with CI; (2) the model constructed by TPOT achieved higher accuracy and lower uncertainty than those by SVM and RF; (3) hyperspectral-based indices correlated more strongly with CI than multispectral ones.