National-scale estimation of nitrogen runoff loss from paddy rice fields in China using machine learning approaches

Accurately estimating total nitrogen (N) runoff loss from rice paddies in China is essential for controlling agricultural non-point source pollution and optimizing N management strategies. However, the effects of key controlling factors and national-scale patterns of N runoff loss remain insufficiently understood. Here, we compiled a comprehensive dataset comprising 1154 records of total N runoff loss from Chinese rice paddies and developed six machine learning models, including eXtreme Gradient Boosting (XGBoost), Random Forest, K-Nearest Neighbors, Support Vector Machine, Multi-Layer Perceptron, and Generalized Additive Model. After Bayesian hyperparameter optimization and nested 10-fold cross-validation, XGBoost achieved the best performance (RMSE = 3.72 ± 0.48 kg ha −1 season −1 , NSE = 0.77). SHapley Additive exPlanations (SHAP) revealed that precipitation, N application rate, soil nitrogen to phosphorus (N:P) ratio, and soil carbon to nitrogen (C:N) ratio served as the dominant controlling factors. National simulations for 2018–2022 estimated an average N runoff intensity of 7.04 ± 3.86 kg ha −1 yr −1 , corresponding to a total load of 212.90 ± 26.53 Gg yr −1 . Single- and double-cropping systems exhibited runoff intensities of 6.64 ± 3.45 and 11.77 ± 5.07 kg ha −1 yr −1 , respectively, and the N runoff load of early rice was 1.57 times that of late rice. High runoff intensities were concentrated along the 30°N belt and southern coastal regions. Hainan Province exhibited the highest intensity (12.15 kg ha −1 yr −1 ), while Hunan and Jiangxi each contributed over 10% of the national N runoff load. Under a 15% N reduction scenario, national N runoff intensity and load were projected to decrease by 0.64 kg ha −1 yr −1 and 19.34 Gg yr −1 , with greater mitigation potential in early rice systems and regions with high N inputs. These findings provide a robust quantitative basis for national-scale N runoff assessment and directly support targeted, optimized N management to mitigate agricultural non-point source pollution.

Authors

Institutions

Publication Details

Journal
Agricultural Water Management
Published
2026-11-18
DOI
https://doi.org/10.1016/j.agwat.2026.110939
Primary Topic
Soil and Water Nutrient Dynamics
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

National-scale estimation of nitrogen runoff loss from paddy rice fields in China using machine learning approaches

Quirijn de Jong van Lier, Zhongen Niu, Ji Qi, Ying Zhao et al.
Agricultural Water Management
Soil and Water Nutrient Dynamics
article

National-scale estimation of nitrogen runoff loss from paddy rice fields in China using machine learning approaches

Quirijn de Jong van Lier, Zhongen Niu, Ji Qi, Ying Zhao, Zhenghui Lv, Yuchen Lv, Jian Liu, Yi Wang, Jiaxin Zhao, Tida Ge, Yong Liu, Jinyi Yu, Pengfei Fan, Libin Shi, Haolin Yang
article en

Abstract

Accurately estimating total nitrogen (N) runoff loss from rice paddies in China is essential for controlling agricultural non-point source pollution and optimizing N management strategies. However, the effects of key controlling factors and national-scale patterns of N runoff loss remain insufficiently understood. Here, we compiled a comprehensive dataset comprising 1154 records of total N runoff loss from Chinese rice paddies and developed six machine learning models, including eXtreme Gradient Boosting (XGBoost), Random Forest, K-Nearest Neighbors, Support Vector Machine, Multi-Layer Perceptron, and Generalized Additive Model. After Bayesian hyperparameter optimization and nested 10-fold cross-validation, XGBoost achieved the best performance (RMSE = 3.72 ± 0.48 kg ha −1 season −1 , NSE = 0.77). SHapley Additive exPlanations (SHAP) revealed that precipitation, N application rate, soil nitrogen to phosphorus (N:P) ratio, and soil carbon to nitrogen (C:N) ratio served as the dominant controlling factors. National simulations for 2018–2022 estimated an average N runoff intensity of 7.04 ± 3.86 kg ha −1 yr −1 , corresponding to a total load of 212.90 ± 26.53 Gg yr −1 . Single- and double-cropping systems exhibited runoff intensities of 6.64 ± 3.45 and 11.77 ± 5.07 kg ha −1 yr −1 , respectively, and the N runoff load of early rice was 1.57 times that of late rice. High runoff intensities were concentrated along the 30°N belt and southern coastal regions. Hainan Province exhibited the highest intensity (12.15 kg ha −1 yr −1 ), while Hunan and Jiangxi each contributed over 10% of the national N runoff load. Under a 15% N reduction scenario, national N runoff intensity and load were projected to decrease by 0.64 kg ha −1 yr −1 and 19.34 Gg yr −1 , with greater mitigation potential in early rice systems and regions with high N inputs. These findings provide a robust quantitative basis for national-scale N runoff assessment and directly support targeted, optimized N management to mitigate agricultural non-point source pollution.

Agricultural Water ManagementVol. 338
Ningbo University (CN), Shenyang Agricultural University (CN), Hunan Normal University (CN), Universidade de São Paulo (BR), Ludong University (CN), Chengdu University of Information Technology (CN), Shandong Normal University (CN), Norwegian Institute of Bioeconomy Research (NO)
Department of Science and Technology of Shandong Province, National Natural Science Foundation of China, Key Technology Research and Development Program of Shandong
Zero hunger, Climate action
Soil and Water Nutrient Dynamics
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon