Robust water–nitrogen optimization for alfalfa in arid regions using APSIM and Gaussian process regression

Water and nitrogen are the key factors limiting alfalfa ( Medicago sativa L.) production in arid and semi–arid regions, yet yield, crop water productivity (WP c ) and agronomic efficiency of nitrogen (AE N ) do not respond consistently to water and nitrogen supply. Based on a three–year field experiment in the Hexi Corridor, a temperate arid continental region of northwestern China, the APSIM–Lucerne model was constructed, calibrated, and then used to generate a wide range of discrete water–nitrogen scenarios, which were then used with a Gaussian process regression (GPR) model to construct continuous surrogate response surfaces for yield, WP c , and AE N . Then, the uncertainty generated by GPR and the knee‑point identification was incorporated into the non–dominated sorting genetic algorithm III (NSGA–III) optimization screening process. This incorporation allowed the selection of high‑confidence management ranges. The results were as follows: (1) APSIM–Lucerne simulated annual yield ( R 2 =0.92) and soil water content ( R 2 =0.85–0.94) accurately; the GPR model surrogates reproduced the APSIM simulations with high accuracy ( R 2 >0.935). (2) Yield decreased with increasing water deficit and was highest at 160 kg·ha –1 nitrogen application rate (N); WP c was highest at 65%–75% of field capacity ( θ f ) with 160 kg·ha –1 ; AE N was highest at 55–65% θ f with 160 kg·ha –1 , so the optimal domains of the three objectives did not coincide. (3) Following NSGA–III optimization and screening by knee point and robustness score, the water–nitrogen management window was determined as an irrigation lower limit of 65.2–71.3% θ f , an upper limit of 70.2–76.3% θ f , and an N of 143.1–161.2 kg·ha –1 . Compared with the yield–maximizing Pareto solution, this window incurred a yield loss of only 3.1–7.0%, while WP c and AE N in c reased by 7.9–10.2% and 6.4–27.9%, respectively. By propagating predictive uncertainty into Pareto–based optimization screening, this study delineates a robust water–nitrogen window and a transferable decision framework for forage production in arid and semi–arid regions.

Authors

Institutions

Publication Details

Journal
Agricultural Water Management
Published
2026-11-20
DOI
https://doi.org/10.1016/j.agwat.2026.110961
Primary Topic
Irrigation Practices and Water Management
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Robust water–nitrogen optimization for alfalfa in arid regions using APSIM and Gaussian process regression

Yayu Wang, Boda Li, Guangping Qi, Yuanbo Jiang et al.
5 links
Agricultural Water Management
Irrigation Practices and Water Management
article

Robust water–nitrogen optimization for alfalfa in arid regions using APSIM and Gaussian process regression

Yayu Wang, Boda Li, Guangping Qi, Yuanbo Jiang, Yanbiao Wang, Yanxia Kang, 银敏华, Haiyan Li
article en

Abstract

Water and nitrogen are the key factors limiting alfalfa ( Medicago sativa L.) production in arid and semi–arid regions, yet yield, crop water productivity (WP c ) and agronomic efficiency of nitrogen (AE N ) do not respond consistently to water and nitrogen supply. Based on a three–year field experiment in the Hexi Corridor, a temperate arid continental region of northwestern China, the APSIM–Lucerne model was constructed, calibrated, and then used to generate a wide range of discrete water–nitrogen scenarios, which were then used with a Gaussian process regression (GPR) model to construct continuous surrogate response surfaces for yield, WP c , and AE N . Then, the uncertainty generated by GPR and the knee‑point identification was incorporated into the non–dominated sorting genetic algorithm III (NSGA–III) optimization screening process. This incorporation allowed the selection of high‑confidence management ranges. The results were as follows: (1) APSIM–Lucerne simulated annual yield ( R 2 =0.92) and soil water content ( R 2 =0.85–0.94) accurately; the GPR model surrogates reproduced the APSIM simulations with high accuracy ( R 2 >0.935). (2) Yield decreased with increasing water deficit and was highest at 160 kg·ha –1 nitrogen application rate (N); WP c was highest at 65%–75% of field capacity ( θ f ) with 160 kg·ha –1 ; AE N was highest at 55–65% θ f with 160 kg·ha –1 , so the optimal domains of the three objectives did not coincide. (3) Following NSGA–III optimization and screening by knee point and robustness score, the water–nitrogen management window was determined as an irrigation lower limit of 65.2–71.3% θ f , an upper limit of 70.2–76.3% θ f , and an N of 143.1–161.2 kg·ha –1 . Compared with the yield–maximizing Pareto solution, this window incurred a yield loss of only 3.1–7.0%, while WP c and AE N in c reased by 7.9–10.2% and 6.4–27.9%, respectively. By propagating predictive uncertainty into Pareto–based optimization screening, this study delineates a robust water–nitrogen window and a transferable decision framework for forage production in arid and semi–arid regions.

Agricultural Water ManagementVol. 338
Gansu Agricultural University (CN)
Zero hunger, Clean water and sanitation
Irrigation Practices and Water Management
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