http

://w

ww

.mdm

.uni

can.

es

The ENSEMBLES Statistical Downscaling Web Portal. End2End Tool for Regional Projection

ACRE Workshop, June 23-25, 2008. Zurich

D. San-Martín, J.M. Gutiérrez, C. Sordo, J. Fernández, D. Frías, M.A. Rodríguez, S.

Herrera, R. Ancell, M. Pons, B. Orfila, E. Díez

Meteorology & Data Mining

Santander Group

Antonio S. Cofiño

http

://w

ww

.mdm

.uni

can.

esMotivation

There are many projects around the world producing global (GCM) and regional (RCM) simulations of climate change.

Many of these projects involve end-uses from impact sectors ...

There is a need of friendly interactive tools so users can easily run interpolation/downscaling jobs on their own data using the existing downscaling techniques and simulation datasets (AR4, Prudence, ENSEMBLES, ...).

However, it is still difficult for end-users to access the stored simulations and to post-process them to be suitable for their own models: daily resolution, interpolation to prescribed locations, etc.

http

://w

ww

.mdm

.uni

can.

es ECHAM5/MPI-OM (200 km)

Statistical Downscaling (SD). WHY?

Typical resolutionof climate change GCMs.

Typical resolution of Seasonal GCMs.

Even if they work at seasonal or climate change scales, end-users normally need daily values interpolated (downscaled) to the local points or grids of interest.

• Mean precipitation• Precip. 90th percentile• Consecutive dry days• Number of heavy events

There is a gap between the coarse-resolution outputs available from GCMs and the regional needs of the end-users.

Surface variables:

http

://w

ww

.mdm

.uni

can.

es

Statistical Downscaling is based on empirical models fitted to data using historical records.

Y = f (X;)

Historical Records

The form and parameters of the model depend of the different tecniques used.

A2

A2

Climatology (1961-90)

GCM Global Predictions

Emission Scenarios

Statistical Downscaling (SD). HOW?

RCMA2 B2Dynamical Downscaling

runs regional climate models in reduced domains with boundary conditions given by the GCMs.

http

://w

ww

.mdm

.uni

can.

esSD. Transfer Functions

Linear Regression: TMaxn = + T850n

Neural Networks: TMaxn = f (T850n ),

Loca

l Rec

ords

of T

max

GCM outputs (closest grid point) 850mb

Linear model

Neural Net.

Neural networks are non-parametric models inspired in the brain.

(T(1ooo mb),..., T(500 mb); Z(1ooo mb),..., Z(500 mb); .......;

H(1ooo mb),..., H(500 mb)) = Xn

Precipitaion,temperatures,

etc.

Yn

http

://w

ww

.mdm

.uni

can.

es

WeatherType

(cluster)

PC1

PC2

Analog set

SD. Weather Types (e.g. analogs)

Pforecast (precip > u) = Ck P(precip > u | Ck) Pforecast(Ck)

frequency mean

http

://w

ww

.mdm

.uni

can.

es

DOWNSCALING HEAVY PRECIPITATION OVER THE UNITED KINGDOM:A COMPARISON OF DYNAMICAL AND STATISTICAL METHODS ANDTHEIR FUTURE SCENARIOS

(HAYLOCK ET AL. 2006)

For some indices and seasons, the spread is very small (e.g. pav in JJA) but for others it is much larger (e.g. pnl90 in DJF). Importantly, for each index the variability among models is of the same order of magnitude as the variability between the two scenarios.

Skill of Statistical Downscaling

The variability of the results obtained using different types of downscaling models in some studies suggests the convenience of using as much statistical downscaling methods as possible when developing climate-change projections at the local scale.

http

://w

ww

.mdm

.uni

can.

eswww.meteo.unican.es/ensembles

30 users from 20 partners

http

://w

ww

.mdm

.uni

can.

esComputing Infraestructure

http

://w

ww

.mdm

.uni

can.

esData Availability. Observations

• ECA (European Climate Assessment & Dataset project). Daily datasets of precipitation, temperature, pressure, humidity, cloud cover, sunshine and snow depth since 1900 over networks of 100-1000 stations.

• Ensembles 50km gridded daily observation records of precipitation and surface temperature. 1950-2006.

http

://w

ww

.mdm

.uni

can.

esData Availability. GCMs

http

://w

ww

.mdm

.uni

can.

esData Availability. Reanalysis & S2D

• NCEP/NCAR Reanalysis1. 1948-2007 • ERA40 ECMWF: 1957-2002 • JRA25 Japanese Reanalysis: 1979-2004

• DEMETER. Multi-model seasonal prediction experiment including seven models ran for six months four times a year using 9 different perturbed initial conditions (9 members).

Data available for the European region:

A smaller worldwide dataset is also available

• ENSEMBLES Stream 1. Check the help in the portal for updated information about this dataset.

Available for Europe

http

://w

ww

.mdm

.uni

can.

esData Availability. ACC

Daily worldwide datasets obtained from different sources:

• CERA• IPCC data centre (PCMDI)• Local Providers.

• PCMDI_CGCM3. Canadian Centre for Climate Modelling and Analysis, including 20th century (from 1951 to 2000) and scenarios A1B, B1 (periods 2046-2065 and 2081-2100).

• CERA_MPI-ECHAM5, including 20th century data (1961-2000) and scenarios A1B, B1, and A2 (2001-2100).

• CNRM-CM3 (local provider), including 20th century (1961-2000) and scenarios A1B, B1, and A2 (2001-2100).

We will continue including datasets as they become available.

http

://w

ww

.mdm

.uni

can.

esData Access Portal

60,Potential Vorticity,PV129,Geopotential,Z130,Temperature,T131,U velocity,U132,V velocity,V133,Specific humidity,Q136,Total Column Water,TCW137,Total Column Water Vapour,TCWV138,Relative vorticity,VO142,Large Scale Precipitation,LSP143,Convective Precipitation,CP151,MSLP,MSL155,Divergence,D157,Relative humidity,R165,10m E-Wind Component,10U166,10m N-Wind Component,10V167,2m Temperature,2T168,2m Dew Point,2D

1000, 925, 850, 700, 500, 300 mb00, 06, 12, 18 , 24 UTC1.125ºx1.125º resolution

http

://w

ww

.mdm

.uni

can.

esData Access: s2d & acc

http

://w

ww

.mdm

.uni

can.

es

Problem: Local climate change prediction for Madrid (Spain): maximum temperature

Goal: Provide daily local values for the summer season june-august 2010-2040 in a suitable format (e.g., text file, or Excel file).

Statistical Downscaling Portal

PrecipitationTemperature

(T(1ooo mb),..., T(500 mb); Z(1ooo mb),..., Z(500 mb); H(1ooo mb),..., H(500 mb))

Xn

Regres, CCA, …Yn = WT Xn Yn

Predictors PredictandsDownscaling Model

This is the structure followed in the portal’s design: predictors + predictand + downscaling method.

Regional zone

Local zone

http

://w

ww

.mdm

.uni

can.

esDemo... My Home

The “My Home” tab allows the user to explore:

1. The zones (pre-defined regions).

2. The profile with the account information.

3. The status of the jobs: queued, running, finished.

http

://w

ww

.mdm

.uni

can.

es

A simple zone with a single predictor parameter:T850mbwas created.

New zones can be easily defined by clicking in the “new zone” button.

Demo... Predictors

http

://w

ww

.mdm

.uni

can.

esDemo... predictand

http

://w

ww

.mdm

.uni

can.

esDemo... Downscaling Method

http

://w

ww

.mdm

.uni

can.

esDemo... Validation

http

://w

ww

.mdm

.uni

can.

esDemo... Regional Projection

http

://w

ww

.mdm

.uni

can.

esDemo...Time to Compute

Five minutes later ...

Scheduling the job

http

://w

ww

.mdm

.uni

can.

esClimate Change Scenarios

Max. Temp

http

://w

ww

.mdm

.uni

can.

esSeasonal validation (RSA)

Precip Tmax

http

://w

ww

.mdm

.uni

can.

es... windows of opportunity with ENSO

Teleconnections with ENSO may bring some seasonal predictability to the extra-tropics and, thus, some window of opportunity for operational seasonal forecast.

http

://w

ww

.mdm

.uni

can.

es... windows of opportunity with ENSO

http

://w

ww

.mdm

.uni

can.

esCurrent Status

• Support for users in ENSEMBLES project.

• Data access for Reanalysis, Seasonal2Decadal and Climate Change models.

• Users can use common observations datasets or upload their own data for downscaling.

• Data access control based on user authentication and authorization.

• User can choose predictors, predictands and transfer function to be used in the downscaling process.

• Quality assessment of the downscaling.

• Download the downscaled data.

http

://w

ww

.mdm

.uni

can.

esFuture actions

• Start to open the tool to wider community outside ENSEMBLES project and Europe.

• Data access and storage: towards remote accessing of datasets based on OPeNDAP. Reanalysis, S2D & ACC simulations.

• Incorporate more statistical downscaling tools.

• Ongoing work on geographically distributed computing and storage based on GRID technologies (EGEE, EELA,…).

Thank you !!!

http://www.mdm.unican.esantonio.cofino@unican.es