Uncertainities in Tipping Warning and Prediction

Uncertainties in Tipping Warning and Prediction Maya Ben-Yami1,2, Andreas Morr1,2, Vanessa Skiba2,3, Lana Blaschke1,2, Sebastian Bathiany1,2, Niklas Boers1,2,4 1 Earth System Modelling, School of Engineering and Design, Technical University of Munich, Munich, Germany 2 Potsdam Institute for Climate Impact Research, Potsdam, Germany 3 Alfred Wegener Institute, Helmholtz Centre for Polar and Marine Research, Potsdam, Germany 4 Department of Mathematics and Global Systems Institute, University of Exeter, Exeter, UK 1. Types of uncertainties 2. AMOC CSD ○Physical false positives ○Observational uncertainties 3. AMOC tipping time 2 Types of uncertainties 1 3 Critical Slowing Down (CSD) As the system approaches the tipping (bifurcation) point: •Variance increases •Autocorrelation increases •Recovery rate from perturbations decreases 4 5 Uncertainties – CSD Need to use indirect proxies, which might not represent the system well enough The observational datasets have uncertainties and biases Need to make assumptions about the dynamical system The driving noise could be non-white Knowledge of the system Measurement of the system Need to use indirect proxies, which might not represent the system well enough Uncertainties – CSD Satellite data for vegetation indices Sea surface temperatures for ocean circulation Ice-core derived melt rates for ice sheet hight 6 Uncertainties – CSD The observational datasets have uncertainties and biases 7 Uncertainties – CSD Quantify the uncertainties Develop methods that account for uncertainties Argue why uncertainties don’t matter use multiple observational datasets together with their uncertainty ensembles calculate restoring rate accounting for autocorrelated noise note CSD without abrupt change is only in very special cases What can we do? 8 AMOC CSD 2 9 AMOC SST Index ASSTI is at the random expected value 🡪 The ASSTI isn’t prone to false positive signs of CSD in the historical period Ben-Yami et al. 2024, EGUsphere 16 SST datasets Ben-Yami et al. 2023, Nat. Comms. 17 SST datasets Ben-Yami et al. 2023, Nat. Comms. 18 SST datasets HadSST4 HadCRUT5 ERSSTv5 Uncertainty ensemble range Ben-Yami et al. 2023, Nat. Comms. 19 SST datasets HadSST4 HadCRUT5 ERSSTv5 Ben-Yami et al. 2023, Nat. Comms. SST datasets HadSST4 HadCRUT5 ERSSTv5 HadSST4 HadCRUT5 ERSSTv5 Distribution of trends Ben-Yami et al. 2023, Nat. Comms. 21 SST datasets HadSST4 HadCRUT5 ERSSTv5 HadSST4 HadCRUT5 ERSSTv5 Ben-Yami et al. 2023, Nat. Comms. median operational product Ben-Yami et al. 2023, Nat. Comms. 24 . . . 25 95% of ensemble trends positive Ben-Yami et al. 2023, Nat. Comms. the system can tip the model is a fold-normal form the noise is white the forcing is linear The MLE method works Modelling assumptions Ben-Yami et al. 2024, Sci. Adv. 32 the system can tip the model is a fold-normal form the noise is white → test methods on a linear model with red noise forcing Modelling assumptions Ben-Yami et al. 2024, Sci. Adv. 33 the system can tip the model is a fold-normal form the noise is white → test methods on a linear model with red noise forcing The MLE method predicts tipping Modelling assumptions Ben-Yami et al. 2024, Sci. Adv. 34 the system can tip the model is a fold-normal form the noise is white the forcing is linear Modelling assumptions Ben-Yami et al. 2024, Sci. Adv. 35 the system can tip the model is a fold-normal form the noise is white the forcing is linear bias to earlier years! Modelling assumptions Ben-Yami et al. 2024, Sci. Adv. 36 the system can tip the model is a fold-normal form the noise is white the forcing is linear Modelling assumptions Ben-Yami et al. 2024, Sci. Adv. 37 the system can tip the model is a fold-normal form the noise is white the forcing is linear bias to earlier years! Modelling assumptions Ben-Yami et al. 2024, Sci. Adv. 38 the system can tip the model is a fold-normal form the noise is white the forcing is linear Modelling assumptions Ben-Yami et al. 2024, Sci. Adv. 39 Sub-polar gyre SSTs – global mean SSTs Sub-polar gyre SSTs – 2x(global mean SSTs) Northern box SSTs – southern box SSTs SPG-1xGMT = SPG-2xGMT = Dipole = AMOC SST fingerprints 40 SST datasets 41 48 of ? 49 50 51 52 53 of ? 54 of ? Models Linear model Ornstein-Uhlenbeck process driven by another Ornstein-Uhlenbeck process + mean trend Fold-normal form Linear or slowing forcing (red noise version replaces dWt with something similar to dUt above) Models 2D Stomell-Cessi model 64 Autocorrelation of last 40 years 65 generate surrogates Autocorrelation of last 40 years 66 generate surrogates modify surrogates 67 Unmodified surrogates Modified surrogates Variance - Mean trend of 1000 surrogates 68 Variance - Mean trend of analysis data 69 Variance - Mean trend of analysis data p<0.05 significance unmodified surrogates 70 Variance - Mean trend of analysis data p<0.05 significance modified surrogates 71 Restoring rate Modified surrogates Analysis trend + significance