Zhu et al. (2026) Effects of soil data resolution on agricultural hydrological simulations in a groundwater-overexploited region: A SWAT and XGBoost-SHAP analysis

Identification

• Journal: Agricultural Water Management
• Year: 2026
• Date: 2026-10-12
• Authors: Jiayao Zhu, Jiangnan Wu, Beibei Ding, Qian Zhang, Xueliang Zhang
• DOI: 10.1016/j.agwat.2026.110849

Research Groups

• College of Land Science and Technology, State Key Laboratory of Efficient Utilization of Agricultural Water Resources, China Agricultural University, Beijing, China

Short Summary

This study evaluates the impact of soil data resolution (horizontal and vertical) on agricultural hydrological simulations using a modified SWAT model in the groundwater-overexploited North China Plain. It finds that a 1 km horizontal resolution with two soil layers offers the best balance for simulating evapotranspiration, groundwater, and crop yield, and that horizontal resolution is more critical than vertical discretization.

Objective

• To evaluate how multiscale soil datasets with varying horizontal resolution and vertical discretization differ in their representation of soil spatial heterogeneity and hydraulic properties within the SWAT modeling framework.
• To quantify the effects of horizontal resolution and vertical profile discretization on SWAT simulation accuracy across evapotranspiration (ETa), groundwater table depth, and crop yield.
• To identify the dominant soil parameters and resolution dimensions driving simulation outcomes through an XGBoost-SHAP surrogate modeling framework, thereby providing process-level insight into the responses of agricultural hydrological simulations to soil data resolution.

Study Configuration

• Spatial Scale: Piedmont plain of Mount Taihang, North China Plain, covering approximately 22,753 square kilometers.
• Temporal Scale: Simulation warm-up period from 1990 to 1992, with the analysis period spanning 1993–2012. ETa calibration/validation from 2002–2004 (SEBAL) and 2000–2011 (MODIS MOD16A2). Groundwater validation from 1993–2010 (national wells) and 2006–2012 (survey wells). Crop yield validation from 1993–2012.

Methodology and Data

• Models used:
  • Modified Soil and Water Assessment Tool (SWAT) model (ArcSWAT2009)
  • XGBoost (eXtreme Gradient Boosting)
  • SHAP (SHapley Additive exPlanations)
  • SUFI-2 algorithm (for SWAT-CUP calibration)
  • SPAW software (for pedotransfer functions)
• Data sources:
  • Soil data: Harmonized World Soil Database v1.0 (10 km, 2 layers), Harmonized World Soil Database v1.2 (1 km, 2 layers), and Institute of Soil Science, Chinese Academy of Sciences (1 km, 5 layers).
  • Meteorological data: 23 weather stations.
  • Evapotranspiration (ETa): SEBAL remote sensing model (1 km x 1 km, monthly) and MODIS MOD16A2 product (1 km, monthly).
  • Groundwater table depth: 162 monitoring wells (16 national, 148 survey).
  • Crop yield: Hebei Rural Statistical Yearbook (county-level winter wheat and summer maize).
  • Other inputs: Digital Elevation Models (DEMs), stream network, land use, agricultural management practices, and groundwater parameters (from previous studies).

Main Results

• Groundwater table depth simulations achieved the highest accuracy across all configurations (Nash-Sutcliffe Efficiency (NSE) of 0.91, coefficient of determination (R²) of 0.91–0.92, Kling–Gupta efficiency (KGE) > 0.94, percent bias (PBIAS) within ±4%), showing least sensitivity to soil data resolution.
• ETa simulations showed intermediate accuracy (R² > 0.75, NSE 0.48–0.55), with the 1 km two-layer dataset (DS-1km-2L) providing the best performance (NSE of 0.55, PBIAS of 0.56%).
• Crop yield simulations exhibited the greatest uncertainty (normalized root mean square error (NRMSE) 18.8–23.1%), with winter wheat yield more responsive to vertical discretization and summer maize yield to horizontal resolution.
• XGBoost-SHAP surrogate models accurately emulated SWAT outputs (test set R² > 0.60 for all targets, up to 0.96 for ETa and groundwater).
• Horizontal resolution improvement (from 10 km to 1 km) systematically increased the importance of soil features (e.g., clay, sand, available water capacity) relative to climate forcing for ETa and crop yield.
• Groundwater dynamics remained predominantly controlled by hydrogeological and climate parameters, irrespective of soil configuration.
• Ablation analysis quantified that refining horizontal resolution from 10 km to 1 km improved surrogate accuracy for all three targets (R² gains of 0.043 for ETa, 0.002 for groundwater, and 0.083 for yield).
• Conversely, adding a fifth vertical layer reduced accuracy for every target (R² reductions of 0.048 for ETa, 0.008 for groundwater, and 0.057 for yield), indicating redundancy and reduced model generalizability.
• The 1 km two-layer dataset (DS-1km-2L) was identified as the optimal soil data configuration for the study region.

Contributions

• Provides a systematic evaluation of the joint effects of horizontal and vertical soil data resolution on coupled agricultural hydrological simulations (ETa, groundwater, crop yield).
• Introduces and applies an interpretable machine learning framework (XGBoost-SHAP) to attribute resolution effects to specific soil parameters, offering mechanistic insights beyond aggregate performance metrics.
• Quantifies that horizontal resolution is the dominant dimension governing simulation accuracy in agricultural hydrological modeling, especially for ETa and crop yield.
• Demonstrates that increasing vertical profile discretization beyond two layers can introduce redundant information and degrade model generalizability rather than improve predictive benefit.
• Offers transferable guidance for prioritizing horizontal over vertical refinement in soil dataset selection for process-based hydrological modeling in groundwater-stressed irrigated regions.

Funding

• National Key Research and Development Program of China (grant number 2023FY101004).

Citation

@article{Zhu2026Effects,
  author = {Zhu, Jiayao and Wu, Jiangnan and Ding, Beibei and Zhang, Qian and Zhang, Xueliang},
  title = {Effects of soil data resolution on agricultural hydrological simulations in a groundwater-overexploited region: A SWAT and XGBoost-SHAP analysis},
  journal = {Agricultural Water Management},
  year = {2026},
  doi = {10.1016/j.agwat.2026.110849},
  url = {https://doi.org/10.1016/j.agwat.2026.110849}
}