Objective
As the U.S. climate becomes more volatile and the population continues
to rise, informed water use is as important as ever.
Myself and a friend of mine, Nathan Jeffries,
set out to further explore urban water use after realizing
that there is limited publicly
available data about city-level water use across the
United States. The motivation behind this project was to fill in geographic data gaps
which may help better inform policy decisions and
improve the use of two vital resources— water and energy.
We aimed to
generate a predictive model for
urban water use, as well as a model able to predict the amount of
electricity needed to process and distribute water,
an important part of the energy-water nexus.
These models fall under the category
of Land Use Regression, or spatial prediction.
We explored four types of models: K Nearest Neighbors, Linear regression,
Ridge regression, and Lasso regression.
Conclusions
We found that KNN
and Ridge models for water use prediction perform the best overall, performing
better for the
larger, higher water consuming cities.
None of our models for processing-electricity
performed well which is
predominately due to the variation in the types of energy
cities use to process their drinking water, be that natural gas, coal, electricity, or others.
The main takeaway from this project is that more primary data
is necessary in order to develop accurate prediction models for urban water use, as well
as processing energy use.
The findings of this project may motivate agencies to prioritize resources towards better
data collection of water use in order to illuminate our society's consumption of this
precious, declining resource.
Hereyou can also check out a poster of this project presented at an
Energy and Resources department forum hosted by UC Berkeley. Please reach out to me if you have any feedback
about this project or would like to collaborate.