A Bootstrap Technique for Testing the Relationship between Local-Scale Radar Observations of Cloud Occurrence and Large-Scale Atmospheric Fields
| Marchand, Roger | Pacific Northwest National Laboratory |
| Beagley, Nathaniel | Pacific Northwest National Laboratory |
| Ackerman, Thomas | DOE/Pacific Northwest National Laboratory |
Limitations in the ability of Global Climate Models (GCMs) to predict clouds create significant uncertainties in predicting and understanding climate. Comparison studies have demonstrated that clouds are among the largest source of uncertainty in global climate model simulations [Cess et al., 1990; Potter and Cess, 2003]. Comparisons of model output and observational data generally require averaging (or aggregating) the observations in an attempt to put them on the same large spatial scale as used by the global models. In the case of ARM ground-based measurements this is particularly troublesome. Even for single point measurements or statistics (e.g. mean values) one must generally assume that temporal averaging of local-scale time-series observations is equivalent to spatial averaging over the model grid cell. Moreover, whereas a Numerical Weather Prediction (NWP) model predicts specific weather events, GCMs predict climate. Thus, one cannot simply ask if the GCM predicts the same cloud field on August 10th as is observed at that time. Rather, one must aggregate the observations over some period of time and analyze to what degree the predicted distribution matches the observed distribution. When a difference is detected, it is difficult to determine the source of the problem (what physical processes or situations are not sufficiently represented by the model) or to determine a corrective action. In this paper we investigate using an atmospheric classification scheme based on fields that are resolved by global climate models (and numerical weather prediction models) as a mechanism to map the large-scale (synoptic-scale) atmospheric state with distributions of local-scale cloud properties. We analyze vertical profiles of cloud occurrence obtained from the vertically pointing millimeter wavelength cloud-radar at the Southern Great Plains site as a function of atmospheric state as determined from a neural network. The goal of the analysis is to evaluate whether or not the profiles of cloud occurrence, when aggregated according to the large-scale atmospheric state, are temporally stable and distinct in a statistically meaningful way. A stable state or class-based map could be of great help in the analysis of GCM predicted cloud properties. By aggregating and comparing model output with observations according to the atmospheric state, a physical context is provided from which to understand any differences between the model output and observations, as well as to separate differences (in total distribution) that are caused by having different weather regimes (or synoptic scale activity) rather than problems in the representation of clouds for a particular regime
This poster will be displayed at the ARM Science Team Meeting.