Parameterization development and evaluation ideally takes a two-step approach (Lohmann et al., 2007). Insight into new processes, and initial parameterization formulation should be guided by theory, process-level observations (laboratory experiments or field studies) or, if these are unavailable, by high-resolution modelling. However, once implemented into large-scale atmospheric models, a thorough testing and evaluation is required in order to assure that the parameterization works satisfactorily for all weather situations and at the scales the model is applied to. Satellite observations are probably the most valuable source of information for this purpose, since they offer a large range of parameters over comparatively long time series and with a very large, to global, coverage. However, satellites usually retrieve parameters in a rather indirect way, and some quantities (e.g., vertical wind velocities) are unavailable. It is thus essential for model evaluation 1. to assure comparability; and, 2. to develop and apply metrics that circumvent the limitations of satellite observations and help to learn about parameterizations. In terms of comparability, the implementation of so-called \"satellite simulators\" has emerged as the approach of choice, in which satellite retrievals are emulated, making use of model information about the subgrid-scale variability of clouds, and creating summary statistics (Bodas-Salcedo et al., 2011; Nam and Quaas, 2012; Nam et al., 2014). In terms of process-oriented metrics, a large range of approaches has been developed, e.g. investigating the life cycle of cirrus from convective detrainment (Gehlot and Quaas, 2012), or focusing on the details of microphysical processes (Suzuki et al., 2011). Besides such techniques focusing on individual parameterizations, the data assimilation technique might be exploited, by objectively adjusting convection parameters and learning about parameter choices and parameterizations in this way (Schirber et al., 2013).In this chapter, we will first introduce the available satellite data, consider their limitations and the approaches to account for these, and then discuss observations-based process-oriented metrics that have been developed so far.

Parameterization development and evaluation ideally takes a two-step approach (Lohmann et al., 2007). Insight into new processes, and initial parameterization formulation should be guided by theory, process-level observations (laboratory experiments or field studies) or, if these are unavailable, by high-resolution modelling. However, once implemented into large-scale atmospheric models, a thorough testing and evaluation is required in order to assure that the parameterization works satisfactorily for all weather situations and at the scales the model is applied to. Satellite observations are probably the most valuable source of information for this purpose, since they offer a large range of parameters over comparatively long time series and with a very large, to global, coverage. However, satellites usually retrieve parameters in a rather indirect way, and some quantities (e.g., vertical wind velocities) are unavailable. It is thus essential for model evaluation 1. to assure comparability; and, 2. to develop and apply metrics that circumvent the limitations of satellite observations and help to learn about parameterizations. In terms of comparability, the implementation of so-called \"satellite simulators\" has emerged as the approach of choice, in which satellite retrievals are emulated, making use of model information about the subgrid-scale variability of clouds, and creating summary statistics (Bodas-Salcedo et al., 2011; Nam and Quaas, 2012; Nam et al., 2014). In terms of process-oriented metrics, a large range of approaches has been developed, e.g. investigating the life cycle of cirrus from convective detrainment (Gehlot and Quaas, 2012), or focusing on the details of microphysical processes (Suzuki et al., 2011). Besides such techniques focusing on individual parameterizations, the data assimilation technique might be exploited, by objectively adjusting convection parameters and learning about parameter choices and parameterizations in this way (Schirber et al., 2013).In this chapter, we will first introduce the available satellite data, consider their limitations and the approaches to account for these, and then discuss observations-based process-oriented metrics that have been developed so far.