-
Uddannelse
Find uddannelse
Studievejledning
Kvalitet
- Forskning
-
Samarbejde
Virksomheder
Kommuner og regioner
Gymnasier
Myndigheder
-
Om AU
Organisation
Kontakt
Om AU
- Organisation
- Ledelse
- Strategi
- AU i tal
- AU's historie
- Internationalt samarbejde
- Bæredygtighed
- Campus 2.0
- Diversitet og ligestilling
- Ledige stillinger
- Presse
- Kontakt
HYDROsim: Fast hydrological simulation for efficient groundwater management
Publikation: Bog/antologi/afhandling/rapport › Ph.d.-afhandling
Numerical groundwater models help us understand the dynamics of the sub-
surface water reservoirs and are essential in our management strategies. The
decision support offered by these models helps decision-makers protect and
optimize the use of the precious fresh drinking water resource. Climate change
introduces a previously unseen stress on these reservoirs making optimal
groundwater management even more vital in the future. Extensive numerical
investigations become necessary for decision-makers to explore all possible
outcomes of an event before making any decisions. Here, an issue presents
itself as the complex numerical models can be computationally heavy to run
making decision support time-consuming and expensive for most purposes.
In this PhD dissertation, we present a method for training a neural network
to replace the numerical model engine in different groundwater management
problems. Initially, we train a neural network to predict drawdown from
groundwater abstraction in two different groundwater models. Later, we turn
to the computationally heavy problem of predicting well recharge areas of wells.
For all problems, results show that the neural networks predict responses with
high accuracy and much faster than the numerical simulation period.
We further contribute with a way of including stochastic model variability
in the neural network predictions using an ensemble of numerical models with
parameter variability. The neural network predicts probabilistic recharge areas
in less than a second. This potentially allows decision-makers to perform
extensive investigations previously deemed infeasible in everyday management
tasks. Furthermore, decisions can now be based on probabilities or risks instead
of deterministic model responses.
We anticipate that the project findings will influence both academia and the
industry, where our research has the potential to benefit future administrative
groundwater management professionals.
surface water reservoirs and are essential in our management strategies. The
decision support offered by these models helps decision-makers protect and
optimize the use of the precious fresh drinking water resource. Climate change
introduces a previously unseen stress on these reservoirs making optimal
groundwater management even more vital in the future. Extensive numerical
investigations become necessary for decision-makers to explore all possible
outcomes of an event before making any decisions. Here, an issue presents
itself as the complex numerical models can be computationally heavy to run
making decision support time-consuming and expensive for most purposes.
In this PhD dissertation, we present a method for training a neural network
to replace the numerical model engine in different groundwater management
problems. Initially, we train a neural network to predict drawdown from
groundwater abstraction in two different groundwater models. Later, we turn
to the computationally heavy problem of predicting well recharge areas of wells.
For all problems, results show that the neural networks predict responses with
high accuracy and much faster than the numerical simulation period.
We further contribute with a way of including stochastic model variability
in the neural network predictions using an ensemble of numerical models with
parameter variability. The neural network predicts probabilistic recharge areas
in less than a second. This potentially allows decision-makers to perform
extensive investigations previously deemed infeasible in everyday management
tasks. Furthermore, decisions can now be based on probabilities or risks instead
of deterministic model responses.
We anticipate that the project findings will influence both academia and the
industry, where our research has the potential to benefit future administrative
groundwater management professionals.
| Originalsprog | Engelsk |
|---|
| Forlag | Aarhus University |
|---|---|
| Status | Afsendt - 31 aug. 2023 |
Se relationer på Aarhus Universitet Citationsformater
ID: 338572528