Stochastic dynamic programming and machine learning under climate change for reservoir and irrigation operations

• Main Author: Mohd. Nasir, Muhammad Adib
• Format: Thesis
• Language: English
• Published: 2022
• Subjects: TA Engineering (General). Civil engineering (General)
• Online Access: http://eprints.utm.my/id/eprint/101430/1/MuhammadAdibMohdNasirPSKA2023.pdf.pdf; http://eprints.utm.my/id/eprint/101430/; http://dms.library.utm.my:8080/vital/access/manager/Repository/vital:151566
• Institution: Universiti Teknologi Malaysia

Anthropogenic climate change potentially causes water shortages over different spatial and temporal scales. Over the coming decades, the impact of climate change would be tangible, with significant increases in the global mean temperature, changes in the frequency and intensity of precipitation, and rising sea levels. These changes will adversely affect the water resource system due to the increased severity of floods, droughts, timing and amount of runoff and evaporation. The reservoir systems require continuous development and revision for optimal operations to deal with the variability of future climate change. Therefore, this study attempted to develop optimal reservoir operation under the new realities of climate change in a tropical agro-hydrological watershed in Perak, Malaysia. Reservoir inflow variability due to climate change affects hydrological processes and irrigation demands at a basin scale. Meta-learning, an ensemble machine learning technique using support vector regression (SVR) and random forest (RF) coupled with the Coupled Model Intercomparison Project Phase 6 (CMIP6) multi-Global Climate Model (multi-GCM), was applied to investigate the impacts of climate change on Kurau River. Five GCMs (CanESM5, MPI-ESM1-2-LR, MRI-ESM2-0, NESM3, and NorESM2-LM) and three shared socioeconomic pathways (SSP1-2.6, SSP2-4.5, and SSP5-8.5) were used. The climate sequences generated by the delta change factor method were applied as input to the meta-learning model to predict the streamflow (reservoir inflow) changes from 2021-2080. The model fitted reasonably well, with Kling–Gupta efficiency (KGE), Nash-Sutcliffe efficiency (NSE), percent bias (PBias), and Root Mean Square Error (RMSE) of 0.79, 0.83, 2.52, and 4.51, respectively, for the training period (1976-1995) and 0.72, 0.72, 5.85, and 6.90, respectively, for the testing period (1995-2005). Future projections of multi-GCM over the 2021-2080 under three SSPs predicted an increase in rainfall for all months except April-June (dry period or off-season), with a higher increase during the wet period (main-season). Temperature projections indicated an increase in maximum and minimum temperatures under all SSPs, with a higher increase of approximately 2.0°C under SSP5-8.5 during 2051-2080 relative to the baseline period of 1976-2005. The model predicted seasonal changes in the inflow by -7.5 to 7.1% and 1.2 to 5.9% during the off-season and the main-season, respectively. A significant inflow decrease was predicted in April and May for all SSPs due to high temperatures during the off-season, with SSP5-8.5 being the worst. The future rice irrigation demand changes for the Kerian Irrigation Scheme compared to the baseline period for two planting periods by -1.0 to 0.1% and -5.3 to -2.6% during the off-season and main-season, respectively. A significant irrigation water demand decrease is predicted in September and October for all SSPs due to increased rainfall during the main-season, with SSP5-8.5 being the most prominent. The stochastic dynamic programming is applied to determine the optimal release policies for Bukit Merah Reservoir considering future climate variability (15 combinations of different GCMs and SSPs for two future periods). The rule curve patterns varied under different scenarios and future periods. The patterns revealed that the reservoir will suffer from tremendous water stress in the far future (2051-2080) than in the near future (2021-2050) and significantly during the off-season.