Intensive Course on Introductory Practical Data Assimilation
Instructor: Takemasa Miyoshi02/08/2010 - 02/10/2010
Laforet Zao Resort & Spa
Supported by Tohoku University
- Students learn fundamental ideas and practical techniques of data assimilation through coding a data assimilation system from scratch by themselves.
- Students develop the Lorenz 40-variable model (Lorenz 1996; Lorenz and Emanuel 1998) using the Runge-Kutta 4th-order scheme and other time-integration schemes.
- Students develop the extended Kalman filter (EKF) based on the Lorenz model.
- As a result, students acquire fundamental understandings of the chaotic nature of the Lorenz system and the Kalman filter procedure for chaos synchronization.
|2/8 Morning||Lecture: Introduction to Data Assimilation and Chaos Synchronization|
|2/8 Afternoon||Exercise: developing and understanding the Lorenz 40-variable model|
|2/9 Morning||Student presentation: the Lorenz 40-variable model
Lecture: How to implement the extended Kalman filter (EKF)
|2/9 Afternoon||Exercise: developing the EKF and 3D-Var|
|2/9 Evening||Lecture: How to implement the ensemble Kalman filter (EnKF)|
|2/10 Morning||Exercise: experiments with EKF and 3D-Var|
|2/10 Afternoon||Student presentation: data assimilation with EKF and comparison with 3D-Var|
- Understanding the behavior of the Lorenz model
Periodic behavior of the Lorenz model with smaller forcing becomes non-periodic and chaotic by increasing the forcing. F=8 indicates chaotic behavior.
(Courtesy of Tsuyoshi Sakai)
- Comparing the Kalman filter and 3D-Var
Kalman filter indicates smaller root mean square errors (RMSE) than 3D-Var.
(Courtesy of Masahiro Sawada)
Debrief by the students
- Tsuyoshi Sakai (Second-year MS student from Tohoku University)
The intensive course of data assimilation was very tough, because it was only three days. But, I could learn the basic theory, and I could also code 3D-Var and Kalman filter with the Lorenz model. I am very happy that my programing skill was improved a lot. And, in this class, I fortunately found a mistake in my MS thesis. :-) Thank you so much for the wonderful class.
- Dr. Masahiro Sawada (Researcher from Tohoku University)
Before spring school on Introductory Practical Data Assimilation, I only knew the word "Kalman filter". After the lecture on fundamental ideas of data assimilation (Kalman filter) and the programing, I realized that it is easier to implement Kalman filter to lorentz model on my own than I thought. I think I could learn the elementary ideas of Kalman filter even though the schedule was a little tight. Through this school, I feel something familiar to Kalman filter as before. It was really a nice and precious experience. I would like to acquire practical ideas of local ensemble Kalman filter and apply initialization/analysis of tropical cyclone. Finally, I would like to express my appreciation to Dr. Miyoshi-san.
- Hiroto Abe (Second-year PhD student from Tohoku University)
I'd been waiting for a chance to study practical data assimilation, because I wanted to code various kinds of computational program. Once I learn one program, it will be helpful to apply this method to other situations. This applications are not always limited to scientific themes, because assimilation techniques such as Kalman filter are besed on concept of optimization problem. That is, acquisition of the techniques will be one step for me to open new world. In this seminer, started from construct of time integral scheme. It took much time to do that because I've not made before then. Next, made program of Kalman filter and confirmed it run normally although a characteristic of the output was a little bit wrong. However, it is important to make program by myself. Attending this seminer encouraged me to learn more data assimilation. I appreciate Dr. Miyoshi and Tohoku university including members of Atmospheric Laboratory to give me such an excited opportunity.
- Dr. Chihiro Tao (Researcher from Tohoku University)
I enjoyed excising data assimilation in the seminar. I could realize through this excise the powerfulness of data assimilation even if the data includes error and/or lack. By expanding this precious opportunity, I would like to continue learning this method with background math/physics and to consider applying it to our research field. I would like to express sincere thanks to the lecturer Dr. Miyoshi-san and everyone having provided this seminar.
Debrief by the instructor
It was a very nice surprise that the students were so keen to learn the practical methods of data assimilation. It was also very nice that the students had various background of not only the atmosphere and ocean of the Earth but also the ionosphere and even Jovian atmosphere.
Although the contents of this class were designed for at least a half-semester graduate course, the students were so enthusiastic for learning and made so much efforts, that most of them succeeded in building and running an extendend Kalman filter. Each student coded their own models and data assimilation systems from scratch. This is really a surprising achievement.
This was really joyful experience for me to have those wonderful outstanding students, and of course, to enjoy the great hotspring in winter Zao.
I am really grateful to Prof. Toshiki Iwasaki and a graduate student Keiichi Mori who made great efforts to make this event happen. I am also very grateful to Tohoku University for their generous support. The Laforet Zao hotel was really helpful to fulfill our various requests such as internet connections and extra rice in lunch and dinner!
Lecture (Courtesy of Guixing Chen)
Dinner time (Courtesy of Guixing Chen)
Kanpai!! (Courtesy of Guixing Chen)
Social (Courtesy of Guixing Chen)
Group photo (Courtesy of Guixing Chen)