Universal Early Warning Signals of Phase Transitions in Climate Systems

This data set contains early warning signal statistics computed from CMIP5 climate simulations used to test the model presented in our paper "Universal Early Warning Signals of Phase Transitions in Climate Systems"

Data is selected and preprocessed as outlined in the paper. For 8 models in the CMIP5 repository, abrupt transitions are identified using the edge detection results of Bathiany et. al. (variables/locations with highest abruptness scores are included, as well as null time series with abruptness scores below a set threshold). Spatiotemporal time data for each of these instances is processed as follows:

• Data provided at monthly resolution is separated into twelve time series sampled annually, in accordance with the processing carried out by Bathiany et. al.
• Runs are truncated such that they end at the time of abrupt transition (or randomly with the same length distribution, for null runs)
• Data is smoothed along the temporal axis using a Gaussian filter with kernel width \(\sigma = 96\) time steps.
• Data is normalized to zero mean and unit variance
• Twelve early warning indicator statistics are computed for each spatiotemporal time series:
  1. Temporal variance
  2. Temporal skewness
  3. Temporal kurtosis
  4. Temporal lag-1 autocorrelation
  5. Temporal lag-2 autocorrelation
  6. Temporal lag-3 autocorrelation
  7. Spatial variance
  8. Spatial skewness
  9. Spatial kurtosis
  10. Spatial distance-1 autocorrelation
  11. Spatial distance-2 autocorrelation
  12. Spatial distance-3 autocorrelation
• Temporal statistics are computed for each spatial grid point (on a sliding window) and then averaged into a single scalar time series
• Spatial statistics are computed separately for each time snapshot

Results are presented in .pkl files formatted for Python Pandas. Each time series ('x' field) has dimension n*12, where n is the length of the time series (up to 600 steps). Other data fields are as follows:

Climate simulations from the CMIP5 collaboration are made available by the World Climate Research Programme and the Program for Climate Model Diagnosis & Intercomparison. Use of this data is subject to the terms outlined at https://pcmdi.llnl.gov/mips/cmip5/terms-of-use.html. Original CMIP5 data can be accessed through the ESGF data portals (see https://esgf-node.llnl.gov/projects/esgf-llnl or the project page at https://esgf-node.llnl.gov/search/cmip5).