Exploring Sources of Uncertainty in Cloud Microphysics Using Statistical Emulation

Microphysical schemes in these scheme models are often complex and interactive. It is necessary to understand the uncertainty before improving their performance for better weather forecast and climate projection. Classical methods for uncertainty and sensitivity analysis require an extremely large number of simulations to obtain robust results when considering multiple parameter inputs simultaneously. This can easily become impractical with a computationally expensive model simulator. The focus of this talk is on describing a statistical emulation approach to overcome such computational barriers, and to facilitate the study of uncertainty in and sensitivity to model inputs describing the microphysical processes for the simulation of a deep convective cloud. Eleven parameters were perturbed within the cloud model, including aerosol concentration, primary freezing, terminal fallspeeds, graupel collection efficiencies, graupel density and ice capacitance. A space-filling experiment design method was used to obtain a set of input combinations on which to base the emulator model, so as to ensure a good coverage of the defined uncertain input space. In this 11-parameter space, only 110 simulations using the complex model simulator were required to construct the emulator model for each cloud response. The emulator provides a statistical representation of the relationship between the uncertain inputs and output of interest, and is much quicker to evaluate than the cloud model itself. Once validated, the emulator can be used in place of the model simulator to obtain the uncertainty and sensitivity measures. We have examined many responses from the cloud model including the specific mass and number concentrations of drops, ice crystal, graupel, the accumulated precipitation, the maximum precipitation rate, updraught and downdraught. Our analysis has enabled the identification of the driving sources of parametric uncertainty for our outputs over the defined uncertain input space under different cloud regimes. For the accumulated precipitation, the main contributors are the fine and accumulation modes of aerosol and the collection efficiency of drops by graupel. For the maximum precipitation rate, the terminal fallspeed of graupel and graupel density also contribute to the uncertainty in addition to the two parameter. It is also found that the interactions in microphysical processes contribute greatly to the uncertainty in some microphysical properties. For example, more than half of the uncertainty in the number concentration of ice crystals is by interaction. Approximately 40% of the uncertainty in graupel number concentration is through interactions. Since graupel particles are a major source of precipitation, it is important to well represent graupel directly and indirectly for better precipitation prediction.