The leading source of the uncertainty in climate projections is the response of clouds to global warming. The grid meshes of current climate models cannot resolve clouds, but advancements in computational power have made kilometer-scale grid spacing possible for regional-scale simulations and will enable next-generation global models to simulate clouds more realistically. However, to reach a more robust estimate of cloud response eventually, new turbulence parameterization schemes need to be developed. Because a grid spacing of a few kilometers puts atmospheric models into the so-called "terra incognita", where clouds and the turbulent eddies associated with them are only partially resolved. This regime differs from those of traditional atmospheric modeling, which assume turbulent eddies are either mostly resolved (in the case of large-eddy simulation) or mostly unresolved (in the case of mesoscale and large-scale modeling). In this presentation, I will introduce the dynamic reconstruction turbulence model (DRM), which employs explicit filtering and reconstruction and allows more active interaction between resolved and unresolved scales. A challenging stratocumulus cloud case will be used to evaluate DRM, and key characters of DRM leading to its success will be discussed. I will also briefly discuss my other research projects related to turbulence parameterization and extreme precipitation.