Application of machine learning techniques to support decision making under uncertainty in water resource management
Application of machine learning techniques to support decision making under uncertainty in water resource management.
McBride, A.and Counsell, C. and Malde, S. and Simm, J.D.
In: AWRA 2018, 4-8 November 2018, Baltimore, USA. (2018)
|Download:|document citation PDF (1362Kb)
|Abstract:||Water companies in the UK are required to produce long-term plans of water resources for their supply area every five years, detailing how they will maintain secure, sustainable supplies , taking account of social and environmental impacts as well as economic costs. Extensive ensemble modelling of water resource systems underpins the production of these reports and the resulting investments chosen to maintain supplies into the future.
Adoption of new guidance on the use of advanced Decision Making Methods (DMMs) and Risk Based Planning has demanded a more comprehensive modelling approach. Modelling and analytical efficiencies are increasingly required for their use and to realise their full benefits. Existing water resources, hydrological, groundwater, and demand models traditionally used by water companies are often not ideally suited for use in these DMMs. Consequently a toolset of approaches is evolving to enable UK water companies to undertake this more complex decision making.
Key elements of this toolset include emulation modelling to complement computationally more expensive process models, machine learning techniques for groundwater assessment and to optimise reservoir control curves considering multiple objectives, and agent based models to explore the spatial and temporal pattern of demand over ensembles of plausible futures. These methods support the rapid simulation times required for applying the DMMs to provide a holistic view of system behaviour under large supply-side, demand-side and policy uncertainties. User-friendly tools and dashboards are being used to explore and communicate the outputs and facilitate effective decision-making, involving all stakeholders.
This toolset of approaches is being increasingly adopted in the UK, demonstrating the potential for innovative methods to interpret and present complex modelling results. Due to the flexible structure of the tools, and the generic approaches used, these techniques can readily be applied to a wide range of settings. However, the absence of physical process representation in some of these methods, and associated implications, must be considered in their application and by planners when interpreting results. Methods in themselves are not a replacement for diligent water planning, but a tool to support it.|
|Item Type:||Conference or Workshop Item (Poster)|
|Uncontrolled Keywords:||Decision making methods, water resources planning, risk based planning, machine learning, United Kingdom|
|Subjects:||Water > General|
|Deposited On:||29 Aug 2018 09:58|
|Last Modified:||14 Nov 2018 06:25|
Repository Staff Only: item control page