Higher improvement in soil health by animal-sourced than plant-sourced organic materials through optimized substitution.

Organic material input can improve soil health and mitigate adverse environmental consequences; however, there is controversy regarding the promoting effects of organic material inputs on soil health due to a lack of quantitative information on the effects of different organic materials from agricultural sources on soil health. This study proposed a novel soil health evaluation method using tree-structured machine learning algorithms to quantitatively assess the effects of animal-sourced (AS) and plant-sourced organic materials (PS) on soil health. The results demonstrated that in the testing set, the two ensemble models, random forest regressor (RFR) and light gradient boosting machine regressor (LightGBMR), showed high coefficient of determination (R 2) of 0.879 and 0.877, respectively, and relative percent deviation (RPD) of 2.83 and 2.84, respectively (P < 0.001). Our findings indicated that the application of AS resulted in superior improvement of soil health for rice, wheat, and maize in comparison to PS. Compared with chemical fertilizer (CF), AS improved soil health indices (SHIs) for rice, wheat, and maize by 23.8–39.8%, PS improved SHIs for rice, wheat, and maize by 13.4–24.4%, and AS + PS improved SHIs for rice, wheat, and maize by 16.0–35.4% (P < 0.001). In addition, the positive effect of AS on SHIs was greater than that of PS. Compared with PS, the SHIs for rice, wheat, and maize increased by 7.34–14.9% for AS (P < 0.001). The reasonable nitrogen (N) application rates for rice, wheat, and maize were 262–290, 235–259, and 299–331 kg ha−1 year−1, respectively. Based on the positive feedback effects on soil health, the reasonable AS substitution rates for rice, wheat, and maize were 26.4–29.2%, 27.0–29.8%, and 29.6–32.8%, respectively. This study confirmed that organic materials improve soil health through optimized substitution, thus contributing to sustainable agricultural management and reducing adverse environmental impacts. [Display omitted] • A novel soil health assessment method using machine learning algorithms was explored. • Manure showed a better effect on soil health indices than plant-sourced materials. • Ideal manure substitution rates for rice, wheat, and maize were 27.8, 28.4, and 31.2%. [ABSTRACT FROM AUTHOR]