[M-GI29] Data assimilation: A fundamental approach in geosciences

[MGI29-P01] Local Particle Filter Experiments with Chaotic Cellular Automata

★Invited Papers

*Ken Furukawa¹, Hideyuki Sakamoto¹, Marimo Ohhigashi, Shin-ichiro Shima², Travis Sluka, Takemasa Miyoshi¹ (1.Institute of Physical and Chemical Research, 2.Graduate School of Simulation Studies, University of Hyogo)

Keywords:data assimilation, particle filter, cellular automata, localization, local particle filter

Data assimilation (DA) has been developed extensively in meteorology. This study explores how DA can be applied to chaotic cellular automata. Cellular automata are useful to model many types of phenomena, including ecology. This study performs a series of local particle filter experiments for two-dimensional three-state chaotic cellular automata called the sheep model. The three states are Sheep, Grass, and Land and are arranged in two-dimensional square grid cells. The state of a cell is evolved by the condition of the adjacent eight cells. The sheep model consists of four processes simulating the ecological dynamics of sheep and grass: sheep eats grass, grass grows, sheep dies, and grass withers. No unique feature about sheep is included in the model. Due to the chaotic nature of the model, if we flip the states of only a few cells, the difference grows rapidly. Therefore, it is difficult to predict the evolution of the sheep model with imperfect initial data. With the local particle filter, we can estimate the true evolution very accurately with imperfect noisy observations. In this presentation, we introduce the sheep model's precise rule and its chaotic behavior. Next, we demonstrate that the local particle filter works well. Finally, we investigate sensitivities to the observation noise and densities in space and time.