Correcting Systematic Bias in the Trend and Frequency Spectrum of Climate Model Simulations - a Wavelets Approach
Abstract
Despite improved understanding of climate physics and computing power, climate models exhibit systematic biases that need to be addressed before their use in impact assessments. Majority of time-based, statistical distribution-based, and frequency-based bias correction approaches can address systematic biases to a limited extent, focussing on a single time scale and overlooking the climate change trend exhibited by the raw data. There is a need for a more generic alternative that can address such biases while mimicking the full extent of variability the observed variable time series exhibits.
Here, we propose a time-frequency-based bias correction approach to correct the statistical attributes of raw climate model simulations by conserving the underlying trend. A discrete wavelet transform (DWT) is employed to disaggregate the climate model time series across dyadic frequency. Our experiment shows that the last decomposition of the approximation wavelet represents the underlying trend in the time series. This allows us to extend our research to use continuous wavelet transform (CWT) to enable a more accurate representation of the underlying spectrum. Similar to DWT, CWT disaggregates time series across spectrum but at a much finer frequency resolution. This capability allows observing and correcting bias in the low frequency variability of the climate model simulation, which a dyadic frequency transform (such as DWT) is unable to do. The robustness of DWT-based (DWBC) and CWT-based (CWBC) bias correction approaches are tested to correct bias in trend, magnitude, and frequency of global mean sea level (GMSL), Arctic sea-ice extent, and sea surface temperature. The results show the improvement of the overall quality of the statistical attributes of the climate model simulations, correcting the bias in trend and preserving the observed variability across spectrum.- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2022
- Bibcode:
- 2022AGUFMGC42L0863K