Downscaling of IPCC scenarios refers to a process of taking global information on climate response to changing atmospheric composition, and translating it to a finer spatial scale that is more meaningful in the context of local and regional impacts. Two general approaches are used in downscaling:

• Dynamical, where a high resolution regional climate model with a better representation of local terrain simulates climate processes over the region of interest.
• Statistical downscaling, where large scale climate features are statistically related to fine scale climate for the region.

The advantage of using dynamical downscaling is that a regional model can simulate local fine-scale feedback processes not anticipated with statistical methods. The disadvantage, however, is that the regional models are far more computationally requiring and that the end performance is highly dependent on the quality of the input data.

Validation of model results
To perform dynamical downscaling of the future climate changes, it is necessary to establish a reliable regional model which simulates well present-day climate for the region of interest. This requires long-term model runs (multi-year, decadal) and comparison with observational data. For description of MM5 model configuration and setup of meteorological scenarios for different case stuies see:

• MM5 model setup, Tehran, Iran;
• MM5 model setup, Korea;
• MM5 model setup: Sisak.
• MM5 model setup, Cyprus;

Model results

• Forecast:   MM5 timeseries for monitoring stations simulated by regular daily forecast for Cyprus region based on global NCEP GFS forecast model from November 2007 and ongoing.
• Reanalysis: simulations are performed based on NCEP FNL (Final) Operational Global Analysis. Complete dataset has been created for year 2008 and additional 5-10 days runs for year 2005.

Climate change simulations approach

• Select one or more IPCC scenarios of interest (A1, A2, B1, B2):
  • Extreme scenarios may be relevant for regional applications
• Select global climate model simulations for the given scenarios according to:
  • Data availability
  • Resolution (spatial and temporal)
  • Climate patterns for the overall region
• Interpolate boundary data from global model to regional grid
  • Horizontally: minimum resolution 27 km, extent of domain 2160x1620 km, approximately 6-8 points on each lateral boundary
  • Vertically: 17 levels, use incremental interpolation if necessary
• Perform regional model simulations
• Compare regional model results with:
  • Present-day simulations (timeseries and patterns)
  • Global predictions downscaled with another method/tool (statistical downscaling)
• Repeat experiments with another input dataset and/or another IPCC scenario
  • Result analysis for multi-scenario multi-model runs

Requirements

• Observational data and a regional model that simulates well present-day climate
• Input data from global climate model dataset:
  • Ideally: resolution 1°x1°, 3 or 6-hourly datasets, 17 vertical levels
• Method/tools for results analysis

Limitations

• Computational Efforts. Running on 8 nodes on a 2* 3 GHz CPU, quad core, 16 GB, 2*450 GB 15K SAS disks, one daily scenario for the Cyprus region takes about 20 minutes. Thus, a one year simulation takes about 5 days of computing time. Ideally, climate change assessment should be performed on a decadal scale using multi- scenario multi model runs (ensembles) to address the ultimately stochastic nature of the problem. This implies the use of large-scale parallel computing or grid computing the meet these requirements.
• Accuracy of input data. The regional model is forced by the global GCM output and inherits the assumptions and errors made in global model simulations. Moreover, coupled AOGCM for the long-term climate simulations have generally lower resolution than the reanalysis models and regular daily forecasting systems and therefore poorer quality of the driving large-scale data compared to the analysis data.