An integration research on the impact of long-term fertilization on soil microbial communities

Abstract

Soils are relevant to our human life and the microbial communities that use them as habitats can actively participate in biogeochemical cycles. Fertilizer application, one of the most common agronomic management practices, is diverse and long-term in nature. However, the effects of long-term fertilization with different types of fertilizers on microbial microorganisms in soils are not fully understood. In this study, we collected bulk soil samples based on 16S rRNA sequencing from 103 publications of 10308 long- term fertilization experiments from various locations worldwide and environmental metadatacorresponding to each sample. To explore the importance of different environmentalvariables as well asthe interaction effects between variables, we evaluated three tree-based machine learning models, RandomForest, XGBoost, and LightGBM, and used the state-of-the-art interpretation method SHAP to interpret the models, whose hyperparameters were optimized by Bayesian optimization algorithm. Ultimately, 20 randomized experiments showed that soil organic carbon, inorganic fertilizer application amount, and sampling depth were the three most essential predictors of soil microbial Shannon diversity. The local SHAP imputation values revealed the robustness of the importance of soil organic carbon, as its SHAP value increased almost monotonically with its value. Furthermore, SHAP analysis for fertilization treatment duration demonstrated that the soil microbial community had reached a steady state under long-term fertilization. In addition, the interaction between the use of N fertilizer and soil organic carbon and soil pH, respectively, was revealed by SHAP interaction analysis. This work demonstrates that the tree-based machine learning algorithm combined with the interpretable machine learning algorithm SHAP has the potential to predict soil microbial Shannon diversity and to analyze global and local attribution. This is critical for capturing the level of environmental factors and directing agricultural operations in a way that preserves soil stability.

Date of Award	Jul 2023
Original language	English
Awarding Institution	University of Nottingham
Supervisor	Jianfeng Ren (Supervisor), Meili Feng (Supervisor) & Jun He (Supervisor)