South New Empirical Downscaling for Africa Bruce Hewitson

  • Slides: 15
Download presentation
South New Empirical Downscaling for Africa ^ Bruce Hewitson & Robert Crane AF 07

South New Empirical Downscaling for Africa ^ Bruce Hewitson & Robert Crane AF 07 Second AIACC Regional Workshop for Africa and Indian Ocean Islands Dakar, Senegal, 24 -27 march 2004

New Empirical Downscaling for Africa Approach Why this approach? Convergent Solutions? Future Directions? Downscaling

New Empirical Downscaling for Africa Approach Why this approach? Convergent Solutions? Future Directions? Downscaling of present rainfall Model Comparison Future Climates

Why Empirical Downscaling? Why this Approach? Little done on sensitivity analysis Little consensus on

Why Empirical Downscaling? Why this Approach? Little done on sensitivity analysis Little consensus on climate requirements Majority of impacts community requires statistics of daily data within month (not actual daily data) Local-scale climate only partially determined by larger-scale controls Requires a stochastic component to the generation of local climate Therefore: GCM fields used to condition a stochastic generation of local climate

Methodology SOM-based Approach -- extracts non-linear relationships in a multivariate data set. Produces: Representations

Methodology SOM-based Approach -- extracts non-linear relationships in a multivariate data set. Produces: Representations of the range and continuum of atmospheric states. Map rainfall distributions to those states

% mm x 10 Percent Frequency of daily rainfall events for two contrasting atmospheric

% mm x 10 Percent Frequency of daily rainfall events for two contrasting atmospheric states % mm x 10

For a given downscaled target location: We have: ~100 atmospheric states, and the rainfall

For a given downscaled target location: We have: ~100 atmospheric states, and the rainfall distribution for each state

To apply the downscaling to the models: We take the GCM fields: Identify corresponding

To apply the downscaling to the models: We take the GCM fields: Identify corresponding atmospheric states for each day in a month Sample from the associated frequency distributions From that, determine dry or wet state and distribution of daily rainfall for the month

Observed Rainfall 1970 -1999 Daily Data NCEP Reanalysis

Observed Rainfall 1970 -1999 Daily Data NCEP Reanalysis

ECHAM Control NCEP Reanalysis

ECHAM Control NCEP Reanalysis

CSIRO Control NCEP Reanalysis

CSIRO Control NCEP Reanalysis

The downscaling procedure still needs some refinement Choice of atmospheric parameters However:

The downscaling procedure still needs some refinement Choice of atmospheric parameters However: