MC3 4 Earth lecture: Kadircan Aksoy (DLR)

We present an information-theoretic framework for estimating carbon transit times in the global carbon cycle using time-lagged mutual information (TLMI). TLMI quantifies statistical dependence between random variables separated by a fixed time lag, providing a way to characterize persistence and turnover of carbon across reservoirs. To estimate TLMI in high-dimensional and nonparametric settings, we employ a family of neural network–based variational methods derived from bounds on KL-divergence called Mutual Information Neural Estimation (MINE). The talk outlines the definition of TLMI, the estimation procedure, and its application to analyzing carbon cycle dynamics.

The guest talk will be streamed via Zoom, please contact Dora Kelemen for access.