Hydropower production is currently the leading renewable energy source in the world.
Other renewables like solar and wind are increasing fast, but a crucial problem
with them is the lack of storage capabilities which makes them highly volatile and
unpredictable in comparison. With water reservoirs you can store the energy
until you need it, which makes it very complimentary to the other sources.
We can even use the energy created from other sources to pump water back up in
the reservoirs for later use. In Norway, hydropower is over 90% of the power
production and it is in turn over 20% of the total hydropower in Europe.
Already this creates a need for good models of the water cycle, but also as more
extreme weather is expected in the years to come, having a good model can help
in better energy planning as well as preventing societal damage from flooding.
Climate scientist predict that there will be more snow on the european
continent in the future, which is also one of the most difficult parts in
predictive hydrology.
We will take use of large, open source datasets from North-America
and northern Europe to classify and group rain catchment areas by properties.
Then we want to see if we can utilize the existing models with more general
parameters so they are catchment property based instead of having a set of
parameters per catchment. We also want to see if we can take knowledge from
what has been learned from fluid dynamics on small scales, for example blood
vessels, and build a model from first principles to use on larger scales like
river networks and underground water flow.
Lastly we will use convolutional neural networks built for video and image
analysis to try and create models that can predict snow melting from images of
mountains in the melting period. These can all be used in combination to improve
the current predictive hydrology as well as prepare for future changes in the
hydrological environment.
I dette prosjektet tar vi utgangspunktet i operasjonelle utfordringer norske krafprodusenter opplever når de skal velge ut hvilke hydrologiske modeller som skal benyttes for å produsere vannkraft. Det eksisterer i dag mange modeller
som brukes til å beskrive prossessene i det hydrologiske kretsløpet, som for eksempel snø, fordamping, grunnvann og topografi. Vi ønsker å studere hvordan parametre i disse modellene påvirker vannføring, da først ved å analysere
usikkerhetene i disse for så å lage forenklede modeller basert på resultatene. Vi vil også lage et verktøy for å hente ut informasjon fra bildeanalyser og værprognoser for å predikere snøsmelting. Målet for en forbedring er i denne sammenheng forstått som økt evne til å forutsi vannføring ved produksjon av vannkraft.