Errors in spectral fingerprints and their effects on climate fingerprinting accuracy in the solar spectrum

- A new approach to use solar spectrum for climate change fingerprinting is introduced.
- The spectral fingerprinting concept is tested using ten years of satellite data.
- Fingerprint errors and their effect on fingerprinting accuracy are estimated.

Using the Earth׳s reflected solar spectrum for climate change fingerprinting is an emerging research area. The spectral fingerprinting approach directly retrieves the changes in climate variables from the mean spectral data averaged across large space and time scales. To investigate this fingerprinting concept, we use ten years of satellite data to simulate the monthly and annual mean reflected solar spectra and the associated spectral fingerprints for different regions over the ocean. The interannual variations in the spectral data are derived and attributed to the interannual variations in the relevant climate variables. The fingerprinting retrieved changes in climate variables are then compared with the actual underlying variable changes from the observational data to evaluate the fingerprinting retrieval accuracy. Two important errors related to the fingerprinting approach, the nonlinearity error and the averaging error in the mean fingerprints, and their impact on the retrieval accuracy, are investigated. It is found that the averaging error increases but the nonlinearity error decreases as the region size increases. The averaging error has minimal effect on the fingerprinting retrieval accuracy in small regions but has more of an impact in large regions. In comparison, the effect of nonlinearity error on the retrieval accuracy decreases as the region size increases. It is also found that the fingerprinting retrieval accuracy is more sensitive to the nonlinearity error than to the averaging error. In addition, we compare the fingerprinting accuracy between using the monthly mean data and the annual mean data. The results show that on average higher retrieval accuracy is achieved when the annual mean data are used for the fingerprinting retrieval.