DA-PAA: A spectral compression method combined with two-dimensional encoding for predicting soil nutrient

Rapid and accurate soil nutrient prediction from visible-near-infrared (Vis-NIR) spectroscopy is important for precision agriculture and soil quality assessment. However, high spectral dimensionality, redundant information, and weak feature representation often limit the performance of existing models. To address these issues, we propose a dynamic adaptive piecewise aggregate approximation (DA-PAA) method for spectral compression and reconstruction. Unlike conventional fixed-segmentation strategies, DA-PAA adaptively adjusts segment lengths according to local spectral variations, reducing dimensionality while preserving critical features. The compressed one-dimensional spectra are then transformed into two-dimensional representations using Gramian Angular Field (GAF) encoding, and a two-dimensional convolutional neural network (2D-CNN) is employed for prediction. Experimental results show that the proposed framework achieves superior performance for soil nutrient estimation. Specifically, it reduces spectral dimensionality from 4200 to 62, substantially improving computational efficiency, while increasing the coefficient of determination (R²) for soil organic carbon and nitrogen prediction by 44.62% and 46.77%, respectively, compared with baseline methods. These results demonstrate that the proposed framework provides an effective and efficient solution for soil nutrient prediction from high-dimensional Vis-NIR spectra.