WCRP-WWRP-THORPEX Model Evaluation and Development Survey - Key Defficiencies
Q1: What are the KEY uncertainties/defficiencis/problems of current models (processes, model physiscs, etc)? What should be evaluated/improved as a priority in models in terms of parameterization and/or interactions among processes?
Climate model uncertiainties:
Aerosol and cloud properties and their impacts on the earth radiation budget
Atmospheric chemistry interactions
Land/atmosphere and ocean/atmosphere interactions
Convective rainfall (timing, climatologic distribution, intraseasonal variability, clouds associated with cumulus convection)
Cloud feedbacks on climate change
Aerosol indirect effects
Uncertainty in regional patterns of precipitation change
Sea ice and icesheet models
Response of plant ecosystems to climate change, including effects of fires and pests
Conservation of energy, tracers and mass in component models.
Divergence between models in polar sensitivity, indicating our lack of understanding of the governing polar processes.
Arctic climate: clouds and stabily stratified boundary layer turbulence.
Land surface modelling (from the perspective of regional, downscaled NWP).
Poor understanding of downward stratopheric coupling and inadequate ozone chemistry in climate models lead to uncertainty in tropospheric (especially high latitute) projections. We do not know what is needed to simulate the effects of stratospheric changes on Arctic surface climate. CCMs do not have interactive oceans and sea ice.
Need to more understanding of stratosphere-troposphere interactions and impact of stratospheric variability on tropospheric climate change - Dall'Amico et al., Clim. Dynamics, 2010.
The representation of the stratosphere - Raleigh friction v. non-orographic gravity wave drag.
Incomplete ability to simulate interannual to multidecadal atmospheric circulation variability features eg NAO and QBO.
What controls the fidelity of simulated ENSO, including amplitude, periodicity and spatial patterns. Many models can simulate ENSO (and other modes of variability such as PDO, AMO, IOD, etc.) but we do not know how to make improvements.
Tropical processes (upper ocean biases, ITCZ and SPCZ, Walker Circulation, South Pacific ocean circulation and pathways to the Equator, SST-deep convection coupling, too diffuse thermocline, etc.), with severe consequences on rainfall teleconnections and future projections.
Systematic error in tropical hydrological cycle. PPT climatologies over ocean are only based on satellite radiances - are some systematic errors due to systematic observational errors?
What is the role of ocean-atmosphere coupling in systematic errors in the Tropics?
How do errors in certain areas affect remote regions and how can these teleconnections be diagnosed?
Understanding the MJO and organized convection and (non-linear) dynamics.
The lack of realistic reproduction of tropical convective transients, especially the MJO and other systematic biases in the Tropics, eg the erroneous double ITCZs, reversed zonal SST gradient in the equatorial Atlantic Ocean and a lack of monsoon rainfall penetration inland.
Uncertainties in parameterizations: model grid scale momentum, heat and water exchange at the ocean -atmosphere interface, sea ice, cumulus properties.
How unresolved processes interact with resolved processes (ie. causes of problems when implementing LES-developped parameterizations into GCMs). How to test and develop parameterizations.
Why is Euro-Atlantic blocking better simulated with high resolution? Can the relevant processes be parameterized?
To what extent is convective behaviour in non-hydrostatic models realistic?
What are the key uncertainties of atmpsheric models? We know the symptoms (lack of MJO, double ITCZ, etc) but what are the causes? What are the uncertainties due to interactions between physics and dynamics, in addition to uncertainties related to dynamical cores (eg advection)?
Understanding the development of errors/deviations on different temporal and spatial scales. The effects of different diagnostic approaches on evaluating a phenomenon, eg differences in cyclone variability and change arising from different climatologies - see unfunded project Intercomparison of Mid-Latitude Storm Diagnostics (IMILAST).
How do we evaluate whether models respond to external forcing realistically?
Low level clouds in subtropics and whether they will decrease or increase with climate change - a key problem in climate sensitivity.
Need to understand the mechanisms by which physiscal parameterization schemes in climate models determine the simulated climate sensitivity, climate variability and climate change signal.
What is the role of diabatic processes and numerics in the development of extra-tropical cyclones?
The absence of the dynamic response of ice sheets in climate models - the dynamic response of ice sheets to warming is the fastest coast-ward motion of ice sheets (~60% of ice loss currently occuring in Greenland - 160gT/yr).
Inadequate record of ice sheet changes and of models to simulate the observed changes.
Representation of ocean circulation in climate models that fail to reproduce eddies.
Neet to better observe and simulate ocean salinity changes.
Inability of observations to adequately describe changes in the Earth's total heat content and its distribution (principally in the ocean) and an inadequately demostrated capability of models to simulate these changes.
Monsoon rainfall - Nearly all AGCMs show poor E. Asian rainfall, though many do have skill in circulation changes (see Zhou references).
Most parameterizations withing probabilistic (ensemble) forecasting systems are deterministic.
For deterministic forecasting key issues are surface initialization and modelling (soil moisture, land characteristics), surface and lower atmosphere coupling (fluxes, treatment of orography, drag and blocking), turbulence in the lower atmosphere (boundary layer schemes, air mass mixing under various stability conditions).
Low cloud fraction (Bony and Durfresne, 2005, GRL)
Cumulus entrainment (Derbyshire et al, 2004, QJRMS)
Effect on diurnal cycle of ppt (Guichard et al, 2004, QJRMS)
Convective anvils (Donner et al, 2001, J. Clim.)
Vertical structure and clouds in extratropical synoptic storms (Trenberth and Fasullo, 2010, J. Clim.)
Nucleation of cirrus clouds and ice supersaturation (Tomkins et al, 2007, QJRMS)
Cloud microphysics in general, including water- and ice-cloud parameterizations, aerosol-cloud interaction, treatment of subgrid scale variations, etc.
Cloud and boundary layer processes should be better observed and validated in models.
Understanding potential predictability of a given component of the climate system before adding complexity to models. How important is the land component for (regional) climate? How much predictability in terms of ppt, radiative forcing, surface climate processes at different timescales (seasonal to decadal).can be attributed to land surface processes? What aspects of land surface processes need to be modelled to attain this predictability?
Atmpsopheric and cloud/convective processes:
Dominance of cloud feedbacks in spread of global climate sensitivty among GCMs
Importance of convection-scheme parameters for this sensitivty - revealed by perturbed physics ensembles
Unrealistic behaviour of GCM convection and rainfall, especially on short timescales.
The importance of convection for model-simulated rainfall characteristics/hydrology impacts, gravity wave generation, ENSO, etc.
The parameterization of deep convection in NWP models (Dx>5km). Many deficiencies are associated with this scheme (the onset of deep convection, the diurnal cycle, the intensity, the propagation, associated precipitation and clouds, etc.). At horizontal scales between 2 and 20km, the parameterization is even more complicated due to the resolved part by the dynamics. The representation of deep convection in cloud resolving models is also an issue. The numerics (advection, numerical diffusion, projection of physical tendencies in dynamical equations) have a much larger impact than in 'large scale' models, and so should be improved. The representation of the PBL is alo crucial for the onset of resolved convection.
Convective cloud parameterization - on which MJO reporducibility critically depends.
In assimilation, need improved observations and intializtion of moisture, clouds and deep convection in a way that is consistent with model physics and dynamics.
Analysis, data-assimilation and prediction of tropopause-level features (Dirren et al., 2003, Martius et al., in press)
Lack of understanding of process interactions, eg. SST-ABL-convection.
The prediction of sea ice thickness, poor prediction of lower tropospheric inversions and clouds in the polar regions.
Aerosol, clouds, ocean heat uptake and transport and ice sheet instabilities.
The level of discrepancy among coupled climate-carbon cycle models, as revealed by the C4MIP project (Friedlingstein et al., 2006) is unacceptable and has serious implications. The models differ as to whether the feedback is trivial or drammatic in terms of understanding the relation between emissions trajectories and CO2 concentrations over the 21st century. These models have been insufficiently validated against essential benchmarks such as the seasonal cycle, interannual variability of atmpspheric CO2 concentration and the Fraction of Absorbed Photosynthetically Active Radiation (fAPAR) at different latitudes. Internally, many models contain demostrably inaccurate representations of key terrestrial carbon cycle processes such as heterotrophic respiration and the response of photosynthesis to temperature and CO2 changes.
The discrepancies among climate models in the strength of the feedback from soil moisture to precipitation, as shown in the GLACE study (Koster and others). As with the carbon cycle, inadequate priority has been given to the comparisons between model predictions and observations for key indicators of surface hydrology such as evapotranspiration and surface temperature. Models often contain parameterizations that are demostrably incorrect, such as representations of evapotranspiration that fail to asymptote as vapor pressure deficit (vpd) increases. A recent analysis (Pitman and Henderson-Sellers) has shown that many models do not even satisfy energy balance constraints at the land surface.