Combining rEW-2DCOS and mechanism-guided adaptive ensemble learning to improve the retrieval of leaf nitrogen, phosphorus, and potassium contents.

Leaf nitrogen, phosphorus, and potassium content (LNC, LPC, LKC) are core nutrient elements and measurable trait parameters essential for assessing vegetation growth status and understanding hydrology-vegetation interactions. However, the spectral characteristics of these elements remain poorly understood, posing a significant challenge for quantitative remote sensing inversion. This study analyzed 303 samples and 21,210 full-spectrum hyperspectral measurements across seven vegetation species, revealing inherent interspecific heterogeneity in their spectra. We quantified spectral heterogeneity using Enhanced Spectral Information Divergence (ESID) and developed a novel r-Enhanced Wavelet Two-Dimensional Correlation Spectroscopy (rEW-2DCOS) method to identify spectral bands exhibiting synergistic correlations with each nutrient element. Validation against traditional CSPA and full-spectrum data confirmed the method's feasibility. The results revealed the density peaks of sensitive bands for LNC (600-860 ​nm, 1230 ​nm, 2080-2250 ​nm), LPC (600-750 ​nm, 1930-2380 ​nm), and LKC (580-830 ​nm, 1680-2350 ​nm). Furthermore, we established a mechanism-guided adaptive ensemble learning regression model (M-AEL) for inversion. The average inversion accuracy (R