Understanding drivers of local lower tropospheric stability

The radiative effect of shallow clouds, especially in the subtropical ocean upwelling regions, is very efficient and mostly controlled by local sea surface temperatures and the lower tropospheric inversion strength. The latter caused the radiative feedback of shallow clouds to switch from positive to negative over the last couple of decades and likely controls cloud feedbacks in the future. Drivers of lower tropospheric inversion strength are not well understood theoretically and vary strongly between reanalyses. We employ convolutional neural networks and explainable artificial intelligence to create maps of drivers of lower tropospheric inversion strength in the subtropical ocean upwelling regions. We quantify the relative relevance of local and remote surface temperatures and find that the edges of regions of deep convection matter much more than their center. The West Pacific Warm Pool is much less and the subtropical Atlantic more relevant than expected. Our results quantify how the theories of tropospheric weak temperature gradient and the convection circus tent play out for setting local free tropospheric temperatures and the local tropospheric inversion stength. Currently, our method is based on large ensemble climate model simulations and the results are robust across models. Applying this framework to observations, it might be possible to constrain the spread of this cloud controlling factor across reanalyses and constrain its future evolution, hence improving projections of the radiative effect of shallow clouds.