heiko paeth heiko.paeth@uni-wuerzburg.de statistical postprocessing of simulated precipitation –...

Post on 19-Dec-2015

215 Views

Category:

Documents

0 Downloads

Preview:

Click to see full reader

TRANSCRIPT

Heiko Paeth heiko.paeth@uni-wuerzburg.de

Statistical postprocessing of simulated precipitation –

perspectives for impact research

IMSC 2010

Heiko Paeth

Institute of Geography, University of Würzburg, Germany

Heiko Paeth heiko.paeth@uni-wuerzburg.de

Diagnosis of model deficienciesannual precipitation totals

Heiko Paeth heiko.paeth@uni-wuerzburg.de

Diagnosis of model deficienciesmonthly precipitation variability

Heiko Paeth heiko.paeth@uni-wuerzburg.de

Diagnosis of model deficienciesPDFs of daily precipitation

climate models:area-mean

precipitation(50km x 50km)

station data:local

information(0,1km x 0,1km)

model datastation data

Heiko Paeth heiko.paeth@uni-wuerzburg.de

Implications for impact research

climate model:

permanentdrizzlingwithin

grid box

hydrological model:

permanent soil moisturization,no peak runoff,

no erosion

Heiko Paeth heiko.paeth@uni-wuerzburg.de

Implications for impact research

climate model:

permanentdrizzlingwithin

grid box

hydrological model:

permanent soil moisturization,no peak runoff,

no erosion

MOS WEGE

Heiko Paeth heiko.paeth@uni-wuerzburg.de

MOS: methodology

MOS

multiplelinear

regressionmodel

cross validation

- 100 iterations with bootstrapping

simulated predictors

- REMO data 1979-2002- rainfall, SAT, SLP, surface wind components

local predictors:max. 0.5° around

each CRU grid cell

EOF predictors:EOFs 1-20 for each

variable

observed predictand

- CRU monthly rainfall 1979-2002

≤ 15 out of 145 predictors are selectedaccording to sig. test

+

Heiko Paeth heiko.paeth@uni-wuerzburg.de

MOS: characteristics

explainedvariance(August)

number ofpredictors

(August)

type ofpredictors

Heiko Paeth heiko.paeth@uni-wuerzburg.de

MOS: resultsannual precipitation totals

Heiko Paeth heiko.paeth@uni-wuerzburg.de

MOS: resultsmonthly precipitation variability

REMO(adj) – CRU(total STD)

REMO - CRU(total STD)

Heiko Paeth heiko.paeth@uni-wuerzburg.de

WEGE: methodology

virtual station rainfall(result)

simulatedgrid-box

precipitation(dynamical part)

local topography(physical part)

v

random distributionin space

(stochastical part)

probability matching

model obs.

Heiko Paeth heiko.paeth@uni-wuerzburg.de

WEGE: results REMO rainfall: - wrong seasonal cycle - underestimated extremes - hardly any dry spells

Weather Generator: - statistical distribution as observed - individual events not in phase with observations

model data

station data

model data postprocessed

original REMO rainfall

rainfall from weather generator

station time series (Kandi)

Heiko Paeth heiko.paeth@uni-wuerzburg.de

WEGE: results

mean daily precipitation intensity mean daily precipitation variability

Heiko Paeth heiko.paeth@uni-wuerzburg.de

Summary

MOS and weather generator worked fine for West Africa and Benin, respectively

impact research in the field of hydrology, agro-economy and heatlh was carried out successfully

MOS approach requires in-phase relationship between model data and observations

weather generator requires high station density with long time series of daily precipitation

top related