Tipping Point Forecasting

This is the official implementation of the code from the paper "Tipping Point Forecasting in Non-Stationary Dynamics on Function Spaces."

Short file descriptions:

Training models on pre-tipping data is performed by running files with names such as nonstationary_lorenz/rno_1d_ode.py or ks/rnn_ks.py. Then, the tipping point analysis is performed in separate analysis Jupyter notebooks or files (e.g., DKW_analysis_[...].ipynb).

Requirements:

• Julia is required to run cloud cover experiments
• Training data for non-stationary Lorenz-63 can be generated with nonstationary_lorenz/run_ode_solver.py.
• Training data for non-stationary KS can be generated with ks/generate_nonstationary_ks.py.
• MixedLayerModel.jl is used to generate training data for the cloud cover experiments
• PyTorch
• Neural Operator Library
• The airfoil dataset can be found on Zenodo