Methodological study demonstrates improved variance estimation in climate attribution, highlighting more reliable uncertainty quantification for regional warming trends.
Key Points
To resolve variance underestimation and improper confidence interval coverage in regularized optimal fingerprinting methods used for climate change detection and attribution.
Formulated the optimal fingerprinting framework as a linear errors-in-variables regression using climate model simulations to estimate internal variability covariance.
Constructed consistent variance estimators using linear shrinkage weight matrices and derived a linearly optimal weight matrix minimizing asymptotic variances.
Evaluated empirical coverage through numerical simulations and applied the estimator to annual mean near-surface air temperature records from 1951 to 2020.
Numerical simulations showed improved empirical confidence interval coverage and reduced interval lengths relative to conventional regularized optimal fingerprinting.
Application to near-surface air temperature data from 1951 to 2020 yielded narrower, more reliable confidence intervals for scaling factors across regional scales.