Understanding and Reducing the Causes of Climate-Model Warm Surface Temperature Biases over the American Midwest: A Multi-Model Evaluation using Data from MC3E

Principal Investigator(s):
Cyril Morcrette, Met Office

We aim to understand and reduce the causes of the warm surface temperature bias over the American mid-west, which is prevalent in a number of climate models (including the American CAM5 model and the British HadGEM3 among others). This will be done using observations collected during the Mid-latitude Continental Convective Cloud Experiment (MC3E) at and around the Southern Great Plains (SGP) facility and comparing them to global climate models. The climate models will be run as if they were weather-forecast models, using analyses as initial conditions. Each model will be set-up to output columns of meteorological data over the SGP site each time-step (so typically with sub-hourly frequency). The surface temperature error, prevalent in a number of climate models suggests that improvement could be made in the manner in which a number of physical processes are represented in those models. A better understanding of the causes of the errors and a better insight into how to fix them is hoped to lead to improvements in a number of global climate models, which will be of benefit for the next round of climate simulations.

In the CAUSES project we wish to apply a new method to assess the role that clouds, and convective clouds in particular, have in contributing to the growth of the surface temperature bias in the mid-west in the models. The new method focuses on the use of high temporal frequency observations and model data to look at the growth of the surface temperature errors on a time-step by time-step basis. This detailed analysis helps to identify which physical process, and hence which parameterization schemes, contributes to the growth of the temperature error. This will help focus model development work. Our analysis method defines 12 cloud regimes consistently in both the model and in the observations. It then calculates the average growth of the model bias in each of the 144 possible model-observation cloud-regime permutations. By studying the cloud properties and surface energy balance in each of the regime-permutations it is also possible to show whether a surface temperature error is due to too much short-wave or long-wave radiation or due to an interaction involving the land-surface. If the error is due to incorrect down-welling radiation the analysis can be extended to determine whether the issue is due to cloud cover, condensed water amount or cloud radiative properties (how cloud properties affect the transfer of short-wave and long-wave radiation). When looking at the short, 5-day timescale, the analysis method developed during our CAUSES pilot-study allowed us to state that the missing of deep convective clouds in HadGEM3 is one of the dominant causes of the surface temperature bias in that model. This information is useful to scientists working on the development of physical parameterizations and the improvement of the GCMs.

Having shown that the method works for 5-day forecasts from HadGEM3, this projects aims to apply the cloud-regime method to other models (including CAM5, WRF, CNRM5) and to study how the error evolves from 5-day to monthly, seasonal and climate timescales. Once we have highlighted the physical processes and cloud-regimes that need improvement, we will work closely with model-development scientists at each institute to suggest parameterization changes. We will then collaborate with them to re-assess any new models versions developed in the light of our recommendations. This focussed assessment of the surface-temperature bias in terms of cloud-regimes is likely to lead to parameterization changes that significantly improve several GCMs.