Cruz et al. (2026) Hydrological and Hydraulic Analysis of Hydropower Plants Reservoirs Under the Influence of Climate Change with a Sequential Machine Learning Model

Identification

• Journal: Water Resources Management
• Year: 2026
• Date: 2026-01-01
• Authors: Josias da Silva Cruz, Leonardo Melo de Mendonça, Nélio Moura de Figueiredo, Claudio Blanco, Antônio Brasil
• DOI: 10.1007/s11269-025-04476-0

Research Groups

• Civil Engineering Graduate Program, Federal University of Pará (PPGEC/ITEC/UFPA), Belém-Pa, Brazil
• School of Naval Engineering, Federal University of Pará (FENAV/ITEC/UFPA), Belém-Pa, Brazil
• Laboratory of Water Engineering and Climate Change, Federal University of Pará (LEHMC/ITEC/UFPA), Belém-Pa, Brazil
• Department of Mechanical Engineering (ENM), University of Brasilia (ENM/TF/UnB), Brasília, DF, Brazil

Short Summary

This study develops a novel sequential cascade machine learning model using Extreme Gradient Boosting (XGBoost) to simulate the hydrological and hydraulic dynamics of the Batalha and Serra do Facão (SEFAC) hydropower plants in Brazil under SSP2-4.5 and SSP5-8.5 climate scenarios, revealing significant future reductions in reservoir storage and outflows, particularly for the downstream SEFAC HPP.

Objective

• To develop and apply a multivariable hydrological modeling methodology based on the Extreme Gradient Boosting (XGBoost) algorithm, using a sequential cascade architecture, to simulate the dynamics (inflow, storage, water level, and outflow) of the Batalha and Serra do Facão (SEFAC) hydroelectric plants under the SSP2-4.5 and SSP5-8.5 climate scenarios, addressing the lack of machine learning models that simulate hydraulic inertia through a causally ordered architecture.

Study Configuration

• Spatial Scale: São Marcos River Basin (drainage area: 7,541.45 km²) in central Brazil, encompassing the Batalha HPP (reservoir: 138.13 km²) and the downstream Serra do Facão (SEFAC) HPP (reservoir: 218.84 km²).
• Temporal Scale:
  • Historical data: Daily period from 2015 to 2024.
  • Future projections: 2025–2050 (near future), 2051–2075 (mid-century), and 2076–2100 (end of century).

Methodology and Data

• Models used: Extreme Gradient Boosting (XGBoost) algorithm with a sequential cascade architecture, developed in Python©.
• Data sources:
  • Hydraulic variables (flow, level, reservoir storage): National Electric System Operator (ONS), daily data (2015–2024).
  • Historical precipitation: Climate Hazards Group InfraRed Precipitation with Station (CHIRPS) dataset.
  • Climate projections: MPI-ESM1-2-LR Global Climate Model (GCM) data from the NASA Earth Exchange Global Daily Downscaled Climate Projections (NEX-GDDP) collection, for Shared Socioeconomic Pathways (SSP) scenarios SSP2-4.5 and SSP5-8.5.

Main Results

• The sequential cascade XGBoost model demonstrated high generalization and predictive accuracy, with Kling-Gupta Efficiency (KGE) values between 0.99 and 1.0 and Root Mean Square Error (RMSE) below 3.2 m³/s for flow simulations.
• Climate projections revealed a critical spatial pattern in the São Marcos River basin: a 10–15% reduction in precipitation upstream of the Batalha HPP versus a 5–10% increase downstream in the SEFAC HPP contribution area.
• For the Batalha HPP, projections indicate a reduction of up to 10% in maximum storage, an 80% reduction in the third quartile of inflow in November, and 20–30% reductions in inflow during the reservoir filling period (January to March).
• The SEFAC HPP is projected to experience more severe impacts, with a 30–40% reduction in storage, maximum future storage consistently below 30% of usable volume, and the third quartile of projected storage for May-September reaching levels close to the historical first quartile (a 48% reduction specifically in July).
• SEFAC HPP outflow shows alarming declines, with an approximate 70% reduction in the third quartile in September and a 38% decrease in average flow.
• Overall, projected reductions of 26–28% in average inflow and 37–39% in average outflow at the SEFAC HPP suggest increasing difficulties in maintaining historical levels of energy production.

Contributions

• Developed an innovative sequential cascade machine learning architecture based on XGBoost that explicitly models the physical causality and interdependence between reservoir state variables (storage, water level, inflow, outflow) by using the model's own simulated outputs as inputs for subsequent predictions, thus capturing the system's hydraulic inertia.
• Demonstrated the robustness and replicability of the proposed methodology for integrated hydrological-hydraulic simulation of complex cascade hydropower systems under climate change scenarios.
• Provided detailed, physically plausible future projections of HPP dynamics under SSP2-4.5 and SSP5-8.5 scenarios, highlighting critical spatial heterogeneity in precipitation impacts and amplified downstream vulnerabilities in cascade systems.
• Reinforced the urgent need for adaptive planning and revision of operational rules for the Brazilian electricity sector to enhance energy security and resilience in the face of climate change.

Funding

• Coordination for the Improvement of Higher Education Personnel-Brazil (CAPES)-Finance Code 001.
• CNPq (research productivity grant, Process 301049/2025-4) for the second author.
• Serra do Facao Energia S.A. (financial support for project ANEEL: PD-06899-0123/2033/SEFAC-CA-024/2023).

Citation

@article{Cruz2026Hydrological,
  author = {Cruz, Josias da Silva and Mendonça, Leonardo Melo de and Figueiredo, Nélio Moura de and Blanco, Claudio and Brasil, Antônio},
  title = {Hydrological and Hydraulic Analysis of Hydropower Plants Reservoirs Under the Influence of Climate Change with a Sequential Machine Learning Model},
  journal = {Water Resources Management},
  year = {2026},
  doi = {10.1007/s11269-025-04476-0},
  url = {https://doi.org/10.1007/s11269-025-04476-0}
}