An Object-Based Approach for Quantification of GCM Biases in the Simulation of Orographic Precipitation.

摘要

An object-based evaluation method to identify and quantify biases of General Circulation Models (GCMs) is introduced. The focus is on how orographic precipitation is simulated by the Eulerian Spectral Transform and the finite volume (FV) dynamical cores within the National Center of Atmospheric Research (NCAR) Community Earth System Model (CESM) with its Community Atmosphere Model (CAM). The “local” biases introduced by dynamical cores and how they evolve with varying model resolution are quantified by looking at simulated precipitation over the Coast Range and the Sierra Nevada mountains on the West Coast of North America. The first step of the object-based method involves identification of orographic precipitation features (study features) simulated differently by the CAM Eulerian Spectral Transform and CAM FV dynamical cores. We examined Atmospheric Model Intercomparison Project (AMIP) model simulations together with Global Precipitation Climatology Center (GPCC) observations to select the study features. CAM FV resembled the observed spatial pattern of precipitation better than the CAM Eulerian Spectral Transform scheme. As the second step of the method, idealized experiments were conducted running the Community Atmosphere Model (CAM) coupled with a simplified physics parameterization to understand the causes of this difference between the CAM FV and the CAM Eulerian Spectral Transform dynamical cores. Three different mechanisms of precipitation were isolated due to (a) stable upslope ascent, (b) local surface fluxes and moisture transport, and (c) resolved downstream waves. The precipitation features related to these mechanisms were isolated as “objects” using pattern recognition methods such as clustering and classification trees. The CAM Eulerian Spectral Transform model simulations become more unrealistic as the resolvable scales of the simulated precipitation gets smaller, and the amount of simulated precipitation gets larger. The reasons of this problematic representation of orographic precipitation by the CAM Eulerian Spectral Transform dynamical core (i.e. bias) can be summarized in three categories: (a) bias due to spectral filtering of the topography, (b) bias in small-scale phenomena due to spectral transform method, (c) grid scale variability (noise) due to spectral transform method. The results also indicated stronger sensitivity of the CAM Eulerian Spectral Transform dynamical core to model resolution.