The Fifth Elephant 2026 Winter Edition

Call for Data Problems and Proposals

Naimish Balaji

@naimish

From Pixels to Equations: Modelling Tropical Cyclone Trajectories with SINDy

Submitted Oct 16, 2026

Tropical cyclones cause huge loss of life and damage every year, and accurate track forecasts are key to early warnings and evacuations. Current forecasts rely on heavy numerical weather prediction models. We explore a lighter, interpretable alternative: using SINDy (Sparse Identification of Nonlinear Dynamics) to discover the governing equations of cyclone motion directly from data. We built a dataset of 1,190 rows from 26 North Indian Ocean cyclones (2007 to 2016), combining IBTrACS best-track positions with wind velocity and wind shear estimated from Meteosat-7 MVIRI satellite imagery (water vapour and infrared channels) using optical flow.

Our model combines two physical relations from the literature, a simple trajectory model (Shan et al., 2020) and a steering-flow/beta-drift model (Emanuel et al., 2006). This guides the SINDy candidate library, and LASSO regression picks out the terms that matter. We compare it against OLS, LASSO, Random Forest and Gradient Tree Boosting.

Comments

{{ gettext('Login to leave a comment') }}

{{ gettext('Post a comment…') }}
{{ gettext('New comment') }}
{{ formTitle }}

{{ errorMsg }}

{{ gettext('No comments posted yet') }}

Hosted by

Jumpstart better data engineering and AI futures