Feb 20, 2012

Statistical Downscaling Method of Global Climate Model

General Circulation Models (GCMs) suggest that rising concentrations of greenhouse gases will have significant implications for climate at global and regional scales. GCMs are restricted in their usefulness for local impact studies by their coarse spatial resolution (typically of the order 50,000 km2) and inability to resolve important sub–grid scale features such as clouds and topography. So, downscaling techniques are used to bridge the spatial and temporal resolution gaps between what climate modelers are currently able to provide and what impact assessors require.
Downscaling Method

Climate model simulations for the 21st century are consistent in projecting precipitation increases in high latitudes and parts of the tropics and decreases in some sub-tropical and lower mid-latitude regions. Outside these areas, the sign and magnitude of projected changes varies between models, leading to substantial uncertainty in precipitation projections. Thus projections of future precipitation changes are more robust for some regions than for others. Projections become less consistent between models as spatial scales decrease. Globally, the negative impacts of future climate change on freshwater systems are expected to outweigh the benefits. By the 2050s, the area of land subject to increasing water stress due to climate change is projected to be more than double that with decreasing water stress. Areas in which runoff is projected to decline face a clear reduction in the value of the services provided by water resources. Increased annual runoff in some areas is projected to lead to increased total water supply.
However, in many regions, this benefit is likely to be counterbalanced by the negative effects of increased precipitation variability and seasonal runoff shifts in water supply, water quality and flood risks. [1]

General assumptions or conventions regarding downscaling of GCMs:
  1.  GCM boundary conditions are the main source of uncertainty affecting all regional downscaling methods
  2.  Statistical and dynamical downscaling has similar skill
  3.  Different downscaling methods yield different scenarios
  4.  There are no universally optimum predictor
  5.  Downscaling extreme events is highly problematic (summer rainfall predictability is very low) 
  6. Skilful downscaling of present climate does not imply skilful downscaling of future scenarios of change

Here are some of findings from IPCC on few environmental aspects:

Sea Level Rise (SLR):
Global mean sea level has been rising and there is high confidence that the rate of rise has increased between the mid 19th and the mid 20th centuries. The average rate was 1.7 ± 0.5 mm/yr for the 20th century, 1.8 ± 0.5 mm/yr for 1961–2003, and 3.1 ± 0.7 mm/yr for 1993–2003.

Precipitation:
Climate projections using multi-model ensembles show increases in globally averaged mean precipitation over the 21st century. The models suggest that precipitation generally increases in the areas of regional tropical precipitation (such as the monsoon regimes) and at high latitudes, with general decreases in the sub-tropics. Increases in precipitation at high latitudes in both the winter and summer seasons are highly consistent across models. Precipitation increases over the tropical oceans and in some of the monsoon regimes, e.g., the south Asian monsoon in summer (June to August) are notable and while not as consistent locally, considerable agreement is found at the broader scale in the tropics.

It is very likely that heavy precipitation events will become more frequent. Intensity of precipitation events is projected to increase, particularly in tropical and high-latitude areas that experience increases in mean precipitation. There is a tendency for drying in mid-continental areas during summer, indicating a greater risk of droughts in these regions. In most tropical and mid-and high-latitude areas, extreme precipitation increases more than mean precipitation.

Global Climate Model (GCM):
To develop spatial downscaling, the use of daily predictors is needed. The predictor variables provide daily information concerning the large-scale state of the atmosphere, while the predictand describes conditions at the site scale (i.e. temperature or precipitation observed at a station). GCM data for the baseline and climate scenario periods are from the various GCMs experiments.
Table 1: List of available predictor variables from NCEP and GCM output
Predictors (Atmospheric Variables)
Origin
Time Window
Mean Sea Level Pressure
NCEP, CGCM1, CGCM2, HadCM3
1961-1990 (base line climate)
Surface Airflow Strength
Surface Zonal Velocity
Surface Meridional Velocity
Surface Vorticity
Surface Wind Direction
Surface Divergence
500hPa Airflow Strength
500hPa Zonal Velocity
500hPa Meridional Velocity
CGCM1, (IS92a)
2000-2099
500hPa Vorticity
500hPa Geopotential
500hPa Wind Direction
500hPa Divergence
850hPa Airflow Strength
850hPa Zonal Velocity
CGCM2, HadCM3 (SRES A2 & B2)
2000-2099
850hPa Meridional Velocity
850hPa Vorticity
850hPa Geopotential
850hPa Wind Direction
850hPa Divergence
Relative or Specific Humidity at 500hPa
Relative or Specific Humidity at 850hPa
Near Surface Relative or Specific Humidity
Mean Temperature at 2m
**NCEP = National Centers for Environmental Prediction; Camp Springs, Maryland
**CGCM = Canadian Global Coupled Model
**HadCM = Hadley Climate Model

HadCM3: HadCM3 is a coupled atmosphere-ocean GCM developed at the Hadley Centre and described by Gordon et al. (1999). It has stable control climatology and does not use flux adjustment. The atmospheric component of the model has 19 levels with a horizontal resolution of 2.5 degrees of latitude by 3.75 degrees of longitude, which produces a global grid of 96X73 grid cells. This is equivalent to a surface resolution of about 417 km X 278 km at the Equator, reducing to 295 km X 278 km at 45° of latitude.

A new radiation scheme is included with 6 and 8 spectral bands in the shortwave and longwave. The radioactive effects of minor greenhouse gases as well as CO2, water vapor and ozone are explicitly represented (Edwards and Slingo, 1996). A simple parameterization of background aerosol (Cusack et al. 1998) is also included.

A new land surface scheme (Cox et al. 1999) includes a representation of the freezing and melting of soil moisture, as well as surface runoff and soil drainage; the formulation of evaporation includes the dependence of stomata resistance to temperature, vapor pressure and CO2 concentration. The surface albedo is a function of snow depth, vegetation type and also of temperature over snow and ice.

A penetrative convective scheme (Gregory and Rowntree, 1990) is used, modified to include an explicit down-draught, and the direct impact of convection on momentum (Gregory et al. 1997). Parameterizations of orographic and gravity wave drag have been revised to model the effects of anisotropic orography, high drag states, flow blocking and trapped lee waves (Milton and Wilson 1996; Gregory et al. 1998).  The large-scale precipitation and cloud scheme is formulated in terms of an explicit cloud water variable following Smith (1990). The effective radius of cloud droplets is a function of cloud water content and droplet number concentration (Martin et al. 1994).

The atmosphere component of the model also optionally allows the transport, oxidation and removal by physical deposition and rain out of anthropogenic sulphur emissions to be included interactively. This permits the direct and indirect forcing effects of sulphate aerosols to be modelled given scenarios for sulphur emissions and oxidants.

The oceanic component of the model has 20 levels with a horizontal resolution of 1.25° X 1.25°. At this resolution it is possible to represent important details in oceanic current structures. Horizontal mixing of tracers uses a version of the Gent and McWilliams (1990) adiabatic diffusion scheme with a variable thickness diffusion parameterization (Wright 1997; Visbeck et al. 1997) is used. There is no explicit horizontal diffusion of tracers. The along-isopycnal diffusivity of tracers is 1000 m2/s and horizontal momentum viscosity varies with latitude between 3000 and 6000 m2/s at the poles and equator respectively.

Near-surface vertical mixing is parametrized partly by a Kraus-Turner mixed layer scheme for tracers (Kraus and Turner 1967), and a K-theory scheme (Pacanowski and Philander 1981) for momentum.  Below the upper layers the vertical diffusivity is an increasing function of depth only. Convective adjustment is modified in the region of the Denmark Straits and Iceland-Scotland ridge better to represent down-slope mixing of the overflow water, which is allowed to find its proper level of neutral buoyancy rather than mixing vertically with surrounding water masses.  The scheme is based on Roether et al. (1994). Mediterranean water is partially mixed with Atlantic water across the Strait of Gibraltar as a simple representation of water mass exchange since the channel is not resolved in the model. The sea ice model uses a simple thermodynamic scheme including leads and snow-cover. Ice is advected by the surface ocean current, with convergence prevented when the depth exceeds 4 m (Cattle and Crossley 1995).

There is no explicit representation of iceberg calving, so a prescribed water flux is returned to the ocean at a rate calibrated to balance the net snowfall accumulation on the ice sheets, geographically distributed within regions where icebergs are found. In order to avoid a global average salinity drift, surface water fluxes are converted to surface salinity fluxes using a constant reference salinity of 35 PSU.

The model is initialized directly from the Levitus (1994) observed ocean state at rest, with a suitable atmospheric and sea ice state. The atmosphere and ocean exchange information once per day. Heat and water fluxes are conserved exactly in the transfer between their different grids.

Downscaling Methods:
Two sets of techniques have emerged as a means of deriving local scale surface weather from regional scale atmospheric predictor variables. Firstly, statistical downscaling is analogous to the model output statistics (MOS). Secondly, Regional Climate Models (RCMs) simulate sub GCM grid scale climate features dynamically using time varying atmospheric conditions supplied by a GCM bounding a specified domain. [2]


Downscaling is justified when variables which are simulated from GCMs are used for impacts. Modelling is unrealistic at the temporal and spatial scales of interest, either because the impact scales are below the climate model’s resolution or because of model deficiencies. However, the host GCM must have demonstrable skill for large scale variables that are strongly correlated with local processes. In practice, the choice of downscaling technique is also governed by the availability of archived observational and GCM data because both are needed to produce future climate scenarios.
Algorithm of Downscaling Method
Data Grid: The predictor variables are supplied on a grid box by grid box basis. On entering the location of your site, the correct grid box will be calculated and a zip file will be made available for download. Each zip file contains three directories:
§  NCEP_1961-2001: This directory contains 41 years of daily observed predictor data, derived from the NCEP reanalyzes, normalized over the complete 1961-1990 period. These data were interpolated to the same grid as HadCM3 (2.5 latitude x 3.75 longitude) before the normalization was implemented.
§  H3A2a_1961-2099: This directory contains 139 years of daily GCM predictor data, derived from the HadCM3 A2 (a) experiment, normalized over the 1961-1990 period.
§  H3B2a_1961-2099: This directory contains 139 years of daily GCM predictor data, derived from the HadCM3 B2 (a) experiment, normalized over the 1961-1990 period.
Grid boxes for the extraction of GCM predictor variable
Bangladesh has fallen in the cell of (25, 25) and (25, 26). The available predictor data then have to correlate with predictand or observed data i.e. temperature or precipitation. After adequate calibration and validation this method/ model will be capable of future data prediction.



Reference:
  1. IPCC technical paper VI, June 2008
  2. Robert L. Wilby and Christian W. Dawson, User Manual of SDSM 4.2, August 2007
  3. Canadian Climate Change Scenarios Network (http://www.cccsn.ca)
  4. The IPCC data distribution center (http://cera-www.dkrz.de/IPCC_DDC/IS92a/HadleyCM3/hadcm3.html)