Projection of temperature and precipitation in Hong Kong in the 21st - - PowerPoint PPT Presentation

Projection of temperature and precipitation in Hong Kong in the 21st century using statistical downscaling

T C Lee T C Lee Hong Kong Observatory

Global climate projections

Global Climate Models / General Circulation Models (GCMs) Human factors

(Greenhouse gases, aerosols, etc.)

Future climate

Intergovernmental Panel on Climate Change (IPCC) Fourth Assessment Report (AR4)

• climate model experiment – 23 models
• climate model experiment – 23 models
• multi-model data set
• 6 greenhouse gas emission scenarios used by

IPCC AR4 in global climate simulations

• from low to high greenhouse gas emissions are

B1, A1T, B2, A1B, A2 and A1FI Low High

Projections of carbon dioxide emission under the six emission scenarios (SRES scenarios) (Source from IPCC)

• 1. Population;
• 2. Economy;
• 4. Environment;
• 5. Equity;
• 6. Globalisation; and
• 7. Climate.

The inset reflects the various assumptions made on the future:

(Source: IPCC)

Why we need to use downscaling ?

Global climate models

• Relatively low spatial resolution (150 – 400 km)
• May not accurately represent local or station level climate

(complex topography, coastal or island locations, etc.) Downscaling is a way to obtain higher spatial resolution output based on GCMs Downscaling technique

Statistical downscaling vs Dynamical downscaling

Statistical downscaling The application of statistical relationships identified in the observed climate, between large and local scale, to GCM outputs Strength

• Station level resolution

Dynamical downscaling Nesting of a finer-scale regional climate model within the coarser global climate model Strength

• High resolution outputs (10km)
• Physically consistent with GCM
• Station level resolution
• Computational inexpensive
• Transferable between regions

Weakness

• Require observational data to

calibrate the downscaling model

• Assume stationary relationship

between predicand and predictors

• Choice of statistical model and

predictors may affect results

• Physically consistent with GCM

Weakness

• Computationally very demanding
• May not be able to transfer between

regions

• Boundary conditions and sub-scale

processes may affect results

Statistical downscaling - basic concept

Statistical downscaling – develop quantitative relationships between large scale predictors and the local predicand Large scale climate

• bservations

(predictors) GCM large scale

• utputs (predictors)

Local scale climate

• bservation

(predictand) Set up statistical relationships (e.g. regression) Downscaling model Downscaled outputs (predicand)

Previous work of the Hong Kong Observatory (HKO)

Temperature projections

• First attempt in 2004 based on IPCC Third Assessment Report

model results

• Update in 2007 based on IPCC AR4 model results (Leung et al.

2007) 2007) Rainfall projections

• First attempt in 2005 based on IPCC Third Assessment Report

model results

• Update in 2009 based on IPCC AR4 model results (Lee et al.

2009)

Data for setting up downscaling model (monthly mean data)

• NCEP re-analysis data / Station observations in southern China
• Hong Kong Observatory Headquarters (HKO Hq) observations

Model data (acquired from IPCC Data Distribution Centre)

• IPCC model monthly mean data
• 16 models and 3 emission scenarios (A1B, B1 and A2)

Data and methodology

Statistical downscaling approach

• Empirical linear regression relationship between the climate of

southern China and that of the HKO (e.g. Average temperature of southern china <>HKO average temperature)

• Urbanization effect was also taken into account for the temperature

projection conducted in 2007. (to simulate the combined effect)

Average temperature of southern China (NCEP reanalysis data) Statistical relationships GCM temperature projection over southern China (predictor) Downscaling model

Schematic diagram showing the downscaling technique for future temperature in Hong Kong

HKO Hq temperatures relationships (linear regression) (regression equation) Projected monthly mean temperatures De-urbanized data HKO Hq

Future urbanization scenarios (a) frozen urbanization (b) continued urbanization

Past and projected annual mean temperature anomaly for Hong Kong (relative to the average of 1980-99)

Low-end - low GHG emission scenario and frozen urbanization High-end - high GHG emission scenario and continued urbanization Middle-of-the-road - average of the GHG emission scenarios as well as of the two situations regarding urbanization

• Average temperature will continue to increase
• More very hot days and less cold days

Average rainfall of southern China (Station rainfall) Statistical GCM rainfall projections over southern China (predictor)

Schematic diagram showing the downscaling technique for future rainfall in Hong Kong

HKO Hq rainfall

• bservation

Statistical relationships (linear regression) Downscaling model (regression equation) Projected monthly rainfall

About 11% increase relative to the 1980-1999 average of 2324 mm About 5% or 120mm less than the 1980-1999 average

• Annual rainfall +11% by the 2090-2099 decade
• -ve decadal rainfall anomaly between 2010 & 2039
• Large variability in decadal rainfall anomaly

Projected occurrence of extremes (rainfall)

Increase in extremely dry years from 2 in 20th century to 4 in 21st century Increase in extremely wet years from 2 in 20th century to 10 in 21st century

Future Work

Projection of temperature and rainfall extremes using a high temporal resolution global climate model data

• Daily multiple (global) model data from Program for Climate Model

Diagnosis and Intercomparison (PCMDI) website

• Multiple linear regression using both surface and upper air

predictors for statistical downscaling

• Extreme indices and GEV analysis of model projections
• Timeframe: Extreme temperature in 2010

Extreme rainfall in 2011

IPCC TAR (2004) IPCC AR4 (2007) Extreme Study (2009/2010) IPCC AR4 Temporal Resolution Monthly data Monthly data Daily data Models 7 16 10 (tentative) Data Volume ~1 G < 10 G ~ 3 T

Comparison between previous and future temperature projections

Scenario B1, A1T, B2, A1B, A2 and A1FI B1, A1B, A2 B1, A1B, A2 Predictands Monthly Tmax, Tmean, Tmin Monthly Tmean Daily Tmax, Tmean, Tmin Predictors Monthly Tmax, Tmean, Tmin Monthly Tmean Daily Surface & upper air (~15 predictors) Downscaling Method Simple linear regression Simple linear regression Multiple stepwise linear regression

THANK YOU

Supplementary Information

A rough estimation of urbanization effect

The magnitude of urbanization effect on temperature is taken as the temperature of the urban station (at HKO Headquarters) minus that of a typical rural station (Ta Kwu Ling) of the region.

• Annual mean temperature difference between HKO Hq and TKL (Tu-r)

for the 18 years data period (1989-2006) is 0.81°C.

• In its early establishment, the HKO Hq was in a countryside setting
• In its early establishment, the HKO Hq was in a countryside setting

surrounded by extensive paddy fields (Doberck 1885). Assume the mean value of Tu-r for the 18 years data period (1885-1902) is “0”.

• The average rate of urbanization between 1885 and 2006 is

computed to be 0.08°C per decade

(Based on Leung et al. 2007)

The future urbanization effect should lie within (i) frozen urbanization (lower bound) (ii) continue at a rate of 0.08oC per decade (upper bound)

Hierarchy of uncertainty in climate projections

GHG emission :

• future GHG emission is uncertain

Large scale GCMs: Large scale GCMs:

• projecting future climate is a challenge
• skill varies from one model to another

Local downscaling:

• extracting local climate details from a

GCM adds uncertainty

Past and projections of rainfall Hong Kong

Time series of the annual rainfall anomaly (with reference to the 1971-2000 average) in Hong Kong from 1950 to 2008. Bold line represents the 9-year running average.

• 800
• 600
• 400
• 200

200 400 600 800 1000 1200 2000-2009 2010-2019 2020-2029 2030-2039 2040-2049 2050-2059 2060-2069 2070-2079 2080-2089 2090-2099 Decade R a i n f a l l a n

• m

a l y ( m m )

er CO2 concentration

A2 A1B 263 mm

• 400
• 200

200 400 600 800 1000 1200 R a i n f a l l a n

• m

a l y ( m m )

157 mm

Higher emissions scenarios projections have larger variance in the forecast

Higher

• 800
• 600
• 400

2000-2009 2010-2019 2020-2029 2030-2039 2040-2049 2050-2059 2060-2069 2070-2079 2080-2089 2090-2099 Decade

B1 325 mm

• 800
• 600
• 400
• 200

200 400 600 800 1000 1200 2000-2009 2010-2019 2020-2029 2030-2039 2040-2049 2050-2059 2060-2069 2070-2079 2080-2089 2090-2099 Decade R a i n f a l l a n

• m

a l y ( m m )

325 mm

Projected changes in mean annual rainfall in HK under A2, A1B, and B1 scenarios The dark line joining the black dots denotes the multi-model ensemble mean

How good are the computer models ?

Simulated annual global mean surface temperature with Hadley Centre model

Source : Alan J. Thorpe, (2005). Climate Change Prediction : A challenging scientific problem, Institute of Physics

Natural + Anthropogenic (Human-made)

Evolution of global climate models

FAR – First Assessment Report SAR – Second Assessment Report SAR – Second Assessment Report TAR – Third Assessment Report AR4 – Fourth Assessment Report (Source : IPCC)