Optimized Estimation of Leaf Mass per Area with a 3D Matrix of Vegetation Indices

Leaf mass per area (LMA) is a key plant functional trait closely related to leaf biomass. Estimating LMA in fresh leaves remains challenging due to its masked absorption by leaf water in the short-wave infrared region of reflectance. Vegetation indices (VIs) are popular variables used to estimate LMA. However, their physical foundations are not clear and the generalization ability is limited by the training data. In this study, we proposed a hybrid approach by establishing a three-dimensional (3D) VI matrix for LMA estimation. The relationship between LMA and VIs was constructed using PROSPECT-D model simulations. The three-VI space constituting a 3D matrix was divided into cubical cells and LMA values were assigned to each cell. Then, the 3D matrix retrieves LMA through the three VIs calculated from observations. Two 3D matrices with different VIs were established and validated using a second synthetic dataset, and two comprehensive experimental datasets containing more than 1400 samples of 49 plant species. We found that both 3D matrices allowed good assessments of LMA (R-2 = 0.76 and 0.78, RMSE = 0.0016 g/cm(2) and 0.0017 g/cm(2), respectively for the pooled datasets), and their results were superior to the corresponding single Vis, 2D matrices, and two machine learning methods established with the same VI combinations.

Automated summary

This study proposed a hybrid approach for estimating LMA by establishing a 3D VI matrix, which was validated on two datasets and achieved good assessment results. The 3D matrices outperformed single VIs, 2D matrices, and two machine learning methods with the same VI combinations.