The proposal suggest using Bayesian networks to build a probabilistic prediction model using historical data.
In this proposal, I suggest using Bayesian networks to build a prediction system that can help decision makers to decide when to send cloud seeding missions.
Bayesian network are probabilistic graphical models that represent a set of random variables and their conditional interdependencies. It's an artificial intelligence representation that make causal assumption explicit, where they were originally developed as a formal means of analysing decision strategies under uncertain conditions.
The proposed research project investigates how Bayesian networks can be implemented using the accumulated historical data of the previous missions, in order to develop a prediction system that can help decision makers and domain experts to decide when to send cloud seeding missions, as well as developing a better understanding of how different factors can affect the cloud seeding outcomes.
Category of the action
Mitigation/Adaptation, Changing public attitudes about climate change
What actions do you propose?
Promoting the importance of using the available data to build a better understanding of how the cloud seeding can be improved.
Who will take these actions?
The proposed project needs a researcher in the field in order to build up the model, analyse the results with the aid of domain experts. Depending on how successful the results are, the project could be further developed to create a user interface that is directly fed by the data from the weather station, and can inform the user about the best time to send the cloud seeding mission.
Where will these actions be taken?
Building the prediction system relies heavily on using the available data. There are several resources that provide data, there's a data that available in R which covers 24 observations from the cloud seeding experiments in Florida:https://vincentarelbundock.github.io/Rdatasets/doc/HSAUR/clouds.html
There's another huge synthesised rainfall dataset that was collected between 1960-2005 in Australia, however it's not publicly available(we have to request for that):
Cloud seeding in the UAE started in 1990, I couldn't find any dataset online about the results of these experiments, but I believe there is some data collected and documented during these years for research purposes, and we have to ask them to give us the data (i'm still not sure about this).
Using local data could help us build more accurate results that are customised for the UAE weather, however using global data to build the model, and comparing it with the local model can help us more in understanding how different factors affect each others in the cloud seeding process.
Also, one of the key advantage of using such a prediction system is the ability to integrate domain experts/knowledge into the model. I attached some resources discussing several ways how this could be achieved.
Another advantage is the ability to scale-up. Unlike machine learning, bayesian networks can be expanded with the introduction of new data, without the need to re-build the model all over again.
What are other key benefits?
Building a prediction model can help decision makers better choose the right timing of sending missions. This could significantly save the costs of sending missions if the chosen timing are accurate (assuming that the model is proven to be highly accurate.
What are the proposal’s costs?
It's hard to estimate the costs as it's a research project.
I would say that it would cost around $150k to hire a computer scientist to build the model. An extra $70k will be needed to build a software that is fed by the weather station data and can produce the results in user friendly manner.
two to three years will be needed to build the model, analyse it and test it.
less than one year is needed to build the software.
Over time, the software is expected to collect more data, this is expected to enhance the prediction model.