Thesis Title: Dynamic Bayesian Networks for Ensemble Seasonal Forecasting

ocelotgiantAI and Robotics

Nov 7, 2013 (3 years and 9 months ago)

81 views

CSAG review: Feb 2008

Name:

Hayley McIntosh

Status:

MSc

Completion date:

20 Feb 2009

Thesis Title: Dynamic Bayesian Networks for Ensemble Seasonal Forecasting

Mini abstract:

Using multiple GCM data and a Dynamic Bayesian Network, this
project is looking to develop a technique to optimally the multiple
ensembles from the different climate models in order to produce
higher resolution seasonal forecasts. Dynamic Bayesian Networks
can merge inputs using probabilities to produce the optimal
combination, the results from which can be merged to produce
different time granularities such as days, weeks or months.

Objectives / Outcomes:

Combine ensemble data into a seasonal forecast
using DBN and hopefully improve the skill coming
from the data.


Current activities / or interesting results:


Coding my prototype for a small area and writing
background chapter. Experimenting with different graph
designs



Workplan for 2008:

Coding the solution. Basic network design finished. Chapter 1 in
progress. Working on finding good data for training from the CHPC
runs. Coding the data manipulation to train distributions in the
DBN. Read more literature on seasonal forecasting in Southern
Africa. Write up methodology (chap 2) when training works. Runs
the inference for a year(s)(?) for results. Compare with current
forecasts for the same year. Add more detail and repeat. Detailed
discussion on the technique and benefits/improvements?/results.
Conclusion and future work to follow.

Milestones:

(explicit,
measurable
, reportable)

End of April: Chapter 1, 2

End of June: Visit to UK (cp.net)

End of Aug: Finished all test runs, compare with Nana’s
skill paper

End of Oct: Results chapter (may take longer, optimistic)

End of Nov: Discussion

Challenges for 2008:

(e.g. skills that need to be
developed, unknowns about data, etc.)

Statistics (data distributions)

Model runs on clusters

Coding (data manipulation)

Partner students: Nana Browne,

First line of help:

Babatunde Abiodun (postdoc), Peter
Johnston

Travel plans / Conferences / Presentations:

UK (May/June)


cp.net/Microsoft

USA (?)