Accurate estimation of Jujube leaf chlorophyll content using optimized spectral indices and machine learning methods integrating geospatial information

Abstract

Leaf chlorophyll content (LCC) is vital for photosynthesis and ecosystem functioning; it influences carbon, water, and energy exchanges while serving as an indicator of photosynthetic activity and nitrogen levels in precision agriculture. Hyperspectral data enable precise LCC monitoring by extracting spectral indices through optimal band combination (OBC) and predicting LCC with machine learning. However, OBC faces dimensionality issues, and machine learning models often overlook geographical influences, potentially reducing prediction accuracy. This study hypothesizes that developing spectral indices from important wavelengths and integrating geospatial data into machine learning models can address these issues and increase prediction accuracy. To test this hypothesis, a framework was developed that first uses elastic net (EN) and the successive projection algorithm (SPA) for wavelength selection, followed by spectral index creation with OBC and ranking with random forest (RF). Support vector regression (SVR), random forest regression (RFR), and geographically weighted least squares support vector regression (GWLS-SVR) were then used to assess the prediction accuracy. Finally, the optimal variables and regression model were identified. The results revealed that the EN- and SPA-based indices had stronger correlations and importance than defined indices. The double-difference index (DDn) and the anti-reflectance index (ARI) are the most robust three-dimensional and two-dimensional spectral indices, respectively. GWLS-SVR requires fewer indices (1–4) to achieve optimal results, with EN-DDn (2R₅₁₉-R₇₇₅-R₉₃₆)-GWLS-SVR performing best (R² = 0.95, RMSE = 0.61, PBIAS = -0.02). This research presents a robust framework with strong adaptability for estimating LCC in a specific study area and region, demonstrating substantial potential for the precise estimation of agroforestry vegetation parameters.