Chucuya et al. (2025) Reconstructing aquifer dynamics with machine learning: Linking irrigation expansion to groundwater decline in a data-scarce hyper-arid region

Identification

• Journal: Agricultural Water Management
• Year: 2025
• Date: 2025-11-29
• Authors: Samuel Chucuya, Roosselvet Pacci, Betzi Bustincio, Edgar Aurelio Taya-Acosta, Wilfredo Alfonso-Morales, Germán Huayna, Edwin Pino-Vargas, Eusebio Ingol-Blanco, Abrahan Mora, Juan Antonio Torres-Martínez, Christian Narvaez-Montoya, Jürgen Mahlknecht
• DOI: 10.1016/j.agwat.2025.110018

Research Groups

• Department of Civil Engineering, Jorge Basadre Grohmann National University, Peru.
• Department of Computer Engineering and Systems, Jorge Basadre Grohmann National University, Peru.
• Department of Geology-Geotechnics, Jorge Basadre Grohmann National University, Peru.
• Perception and Intelligent Systems (PSI) Research Group, School of Electrical and Electronics Engineering, Universidad del Valle, Colombia.
• Civil Engineering Department, New Mexico State University, USA.
• Escuela de Ingeniería y Ciencias, Tecnologico de Monterrey, Mexico.

Short Summary

This study utilizes machine learning (BPNN) to reconstruct fragmented groundwater records in the hyper-arid Caplina aquifer, revealing a 0.6 m/yr water table decline driven by a 400% expansion of irrigated agriculture over three decades. The research highlights the critical role of seawater intrusion in maintaining stable water levels near the coast while severely degrading water quality.

Objective

• To reconstruct missing groundwater level observations and identify spatiotemporal dynamics in a data-scarce, hyper-arid transboundary aquifer to link irrigation expansion with groundwater depletion and seawater intrusion.

Study Configuration

• Spatial Scale: Caplina aquifer and basin, Peru–Chile border, Northern Atacama Desert (approx. 5,190 km²).
• Temporal Scale: 2002–2022 (21 years of groundwater records); 1989–2022 for irrigation expansion analysis.

Methodology and Data

• Models used: Back-propagation neural networks (BPNN), Multivariate Imputation by Chained Equations (MICE), missForest, k-nearest neighbors (KNN), K-means clustering (with Soft-DTW), and Random Forest (RF) for supervised land-use classification.
• Data sources: In-situ groundwater level records from 119 production wells (National Water Authority of Peru - ANA), Landsat 5 ETM and Landsat 8 OLI/TIRS satellite imagery, and hydrochemical data (Chloride concentrations).

Main Results

• Imputation Accuracy: BPNN was the most effective method for filling data gaps, achieving the lowest average errors (MAE = 0.60 m; RMSE = 1.01 m), particularly at high missingness levels (up to 50%).
• Aquifer Dynamics: Clustering revealed three distinct behaviors: a sharp decline in the central aquifer (0.6 m/yr), stable levels in coastal areas, and stationary trends in urban valley zones.
• Irrigation Expansion: Irrigated areas increased by 406%, from 5,381.5 ha in 1989 to 27,246.3 ha in 2022. This expansion is strongly correlated with groundwater extraction volumes (r² = 0.89).
• Water Quality Degradation: While coastal water levels appeared stable, chloride (Cl⁻) concentrations rose from 1,996 mg/L in 2002 to 6,420 mg/L in 2022, confirming intensive seawater intrusion.
• Water Balance: The aquifer faces a severe deficit, with an estimated negative balance of -177 hm³/yr (extraction of 197 hm³/yr vs. recharge of ~20 hm³/yr).

Contributions

• First application of machine learning imputation and clustering to analyze spatiotemporal groundwater variability in a coastal aquifer within a hyper-arid zone (<10 mm/yr precipitation).
• Development of an end-to-end workflow coupling statistical learning and remote sensing that is transferable to other data-scarce dryland basins.
• Provides quantitative evidence of the failure of existing groundwater management policies and the urgent need for alternative water sources (e.g., desalination) in the Atacama Desert.

Funding

• National University Jorge Basadre Grohmann (UNJBG), Tacna, Peru.
• AGUA-H2O-UNJBG team and the Investigation Institute of the UNJBG.
• Research project: "Determinación de las reservas de agua dulce del acuífero Caplina y evaluación de riesgos de la salinización para un manejo sostenible del agua subterránea, Tacna, Perú" (Reference code: RR 15241–2025-UNJBG).

Citation

@article{Chucuya2025Reconstructing,
  author = {Chucuya, Samuel and Pacci, Roosselvet and Bustincio, Betzi and Taya-Acosta, Edgar Aurelio and Alfonso-Morales, Wilfredo and Huayna, Germán and Pino-Vargas, Edwin and Ingol-Blanco, Eusebio and Mora, Abrahan and Torres-Martínez, Juan Antonio and Narvaez-Montoya, Christian and Mahlknecht, Jürgen},
  title = {Reconstructing aquifer dynamics with machine learning: Linking irrigation expansion to groundwater decline in a data-scarce hyper-arid region},
  journal = {Agricultural Water Management},
  year = {2025},
  doi = {10.1016/j.agwat.2025.110018},
  url = {https://doi.org/10.1016/j.agwat.2025.110018}
}

Generated by BiblioAssistant using gemini-3-flash-preview (Google API)