Non-negligible effects of cloud vertical overlapping assumptions on longwave spectral fingerprinting studies

In order to monitor and attribute secular changes from outgoing spectral radiances, spectral fingerprints need to be constructed first. Large-scale model outputs are usually used to derive such spectral fingerprints. Different models make different assumptions on vertical overlapping of subgrid clouds. We explore the extent to which the spectral fingerprints constructed under different cloud vertical overlapping assumptions can affect such spectral fingerprinting studies. Utilizing a principal component-based radiative transfer model with high computational efficiency, we build an OSSE (Observing System Simulation Experiment) with full treatment of subgrid cloud variability to study this issue. We first show that the OLR (outgoing longwave radiation) computed from this OSSE is consistent with the OLR directly output from the parent large-scale models. We then examine the differences in spectral fingerprints due to cloud overlapping assumptions alone. Different cloud overlapping assumptions have little effect on the spectral fingerprints of temperature and humidity. However, the amplitude of the spectral fingerprints due to the same amount of cloud fraction change can differ as much as a factor of two between maximum random versus random overlap assumptions, especially for middle and low clouds. We further examine the impact of cloud overlapping assumptions on the results of linear regression of spectral differences with respect to predefined spectral fingerprints. Cloud-relevant regression coefficients are affected more by different cloud overlapping assumptions than regression coefficients of other geophysical variables. These findings highlight the challenges in constructing realistic longwave spectral fingerprints and in detecting climate change using all-sky observations.