Machine learning (ML) models offer advantages over process-based models for real-time reservoir operation modelling, yet the impact of input variable selection (IVS) and data pre-processing on model performance remains underexplored. This study investigates various input variables for simulating daily reservoir outflow, using the Sirikit reservoir in Thailand as a case study. The datasets include daily Sirikit storage and inflow, outflow of Bhumibol (neighbouring reservoir), downstream discharge, and temporal factors (month and day of the week). Time series decomposition and correlation analyses were used to assess data relationships. We tested seven ML models: multiple linear regression, support vector machine, K-nearest neighbour, classification and regression tree, random forest, multi-layer perceptron, and recurrent neural network (RNN). The optimal input set comprised the previous day’s storage, inflow from 2 days before to 2 days after, and month. With these inputs, all ML models simulated outflow adequately (KGEtraining = 0.42–1.0 and KGEtesting = 0.46–0.56), with RNN showing the most potential for improvement. Input scaling significantly enhanced model performance, reducing RMSEtraining by 44 m3 s-1 and RMSEtesting by 14 m3 s-1. This study’s novelty lies in its comprehensive insights of IVS and data scaling, highlighting their critical roles in enhancing ML model application for operational reservoir simulations.@en

Study region: Upper region of the Greater Chao Phraya River (GCPR) basin in Thailand. Study focus: The upper GCPR basin is highly regulated by multipurpose reservoirs, which obviously have altered the natural streamflow. Understanding quantitative effects of such alteration is crucial for effective water resource management. Therefore, this study aims to assess how reservoir operation affects the water balance, daily flow regime and extreme flows in this basin. For this purpose, we reconstructed streamflow in the naturalized (no reservoir) and baseline operation scenarios using the (∼1 km resolution) distributed model. To overcome data scarcity, we ran the model with global data and parameterization. A target storage-and-release-based reservoir operation module was applied in the baseline operation scenario. The model results were analyzed in comparison to observations in a wet year, a dry year, and the period 1989–2014. New hydrological insights for the region: The reservoir operation resulted in more evaporation. It inverted the natural flow seasonality and smoothed the daily flow regime with decreasing high flows, increasing mean flows and low flows, greater baseflow contribution, and lower flashiness. It prevented or mitigated many historical extreme flow incidents. The annual flood peaks and minimum flows were markedly mitigated in terms of both magnitudes and frequencies, but their timing became more variable and difficult to predict. Altogether, the results highlighted the importance of effective decision making for real-time operation, which remain challenging in practice.

@en

Study region Upper region of the Greater Chao Phraya River (GCPR) basin in Thailand. Study focus This study presents a (∼1 km resolution) distributed hydrological model, wflow_sbm, with global spatial data and parameterization for estimating daily streamflow in the upper GCPR basin, with the aim to overcome in situ data scarcity often occurring in Southeast Asia. We forced the model with the MSWEP V2 precipitation and eartH2Observe potential evapotranspiration datasets. Seamless distributed parameter maps based on pedotransfer functions (PTFs) and literature review were applied to bypass calibration. Only the KsatHorFrac parameter determining the lateral subsurface flow was calibrated. A target storage-and-release-based reservoir operation module (ROM) was implemented to simulate reservoir releases. We compared the simulated daily streamflows obtained from different PTFs and evaluated the model performance in the period 1989–2014. New hydrological insights for the region The global-data-driven wflow_sbm model can reconstruct daily streamflow in the upper GCPR basin, especially for natural catchments (KGE = 0.78). The ROM can capture the seasonal variability of reservoir releases, but not very accurately at the daily timescale (KGE = 0.43) since the actual reservoir operations are too complex. Different PTFs and KsatHorFrac values only introduce little uncertainty in the streamflow results. Therefore, the proposed model provides an opportunity for streamflow estimation in other ungauged or data-scarce basins in Southeast Asia. Nonetheless, the difficulty in the reservoir system modeling reflects the necessity of better understanding of human intervention on daily streamflow.@en