Evolution of paddy soil fertility in a millennium chronosequence based on imaging spectroscopy

It is essential for future security and soil sustainability to understand how soil fertility evolves under long-term anthropogenic influences. This study investigated the feasibility of laboratory-based imaging spectroscopy (IS) techniques to predict fertility properties (soil organic matter (SOM), total nitrogen (TN), total phosphorus (TP), total potassium (TK), pH) of soil profiles at different ages of rice planting. Three methods, support vector machine (SVM), partial least square regression (PLSR) and back propagation neural network (BPNN), were used to calibrate the models. The prediction performances of full-spectrum multivariate models and variable-selected (i. e., the bootstrapping soft shrinkage (BOSS) algorithm) optimization models were compared. The best models were used to map the fertility properties for every profile and the synthetic index for soil fertility of profiles were computed and used to quantify the soil fertility. And then the evolution of soil fertility was analyzed on a millennium scale. The results showed that the five fertility properties could be predicted and mapped for all profiles, and the BOSS-SVMR models performed better than other models. According to the independent validation results, SOM, TN, and pH were predicted well (R2 >= 0.82, RPIQ >= 3.37), and the prediction performance for TP and TK was acceptable (0.50 < R2 < 0.66, 2.02 < RPIQ < 2.70). The vertical distribution of soil fertility at different rice-planting ages showed that the soil fertility in the surface layer (0-30 cm) was high and decreased gradually with depth, tending to be stable in the deeper layer (30-100 cm). Rice planting significantly improved surface soil fertility. The linear function fit the changes in soil fertility with planting age better than the power function, indicating that the soil fertility in this area may have a high capacity for improvement. It was concluded that IS technology could be applied to high-resolution mapping of fertility properties in the intact paddy soil profiles. These results can offer new insights on soil fertility evolution and future soil sustainability under longterm anthropogenic influences.